refactor(chat): 拆分聊天逻辑至独立 ChatService 并提取 LLM 配置

将 ChatController 中的聊天与向量搜索流程整体迁移到 ChatServiceImpl，
新增 AppConfig.LLmConfig 集中管理系统提示语与最大消息长度，
消除控制器层复杂逻辑，提升可维护性与配置动态化能力。

src/main/java/com/yolo/keyborad/controller/ChatController.java

@@ -1,6 +1,5 @@
 package com.yolo.keyborad.controller;
 
-import cn.dev33.satoken.context.mock.SaTokenContextMockUtil;
 import cn.dev33.satoken.stp.StpUtil;
 import cn.hutool.core.util.IdUtil;
 import cn.hutool.core.util.StrUtil;
@@ -12,29 +11,19 @@ import com.yolo.keyborad.mapper.QdrantPayloadMapper;
 import com.yolo.keyborad.model.dto.chat.ChatReq;
 import com.yolo.keyborad.model.dto.chat.ChatSaveReq;
 import com.yolo.keyborad.model.dto.chat.ChatStreamMessage;
-import com.yolo.keyborad.model.entity.KeyboardCharacter;
-import com.yolo.keyborad.model.entity.KeyboardUserCallLog;
-import com.yolo.keyborad.service.KeyboardCharacterService;
-import com.yolo.keyborad.service.KeyboardUserCallLogService;
+import com.yolo.keyborad.service.ChatService;
 import com.yolo.keyborad.service.impl.QdrantVectorService;
 import io.qdrant.client.grpc.JsonWithInt;
 import io.swagger.v3.oas.annotations.Operation;
 import io.swagger.v3.oas.annotations.tags.Tag;
 import jakarta.annotation.Resource;
 import lombok.extern.slf4j.Slf4j;
-import org.springframework.ai.chat.client.ChatClient;
-import org.springframework.ai.openai.OpenAiChatOptions;
 import org.springframework.ai.openai.OpenAiEmbeddingModel;
 import org.springframework.http.codec.ServerSentEvent;
 import org.springframework.web.bind.annotation.*;
 import reactor.core.publisher.Flux;
-import reactor.core.publisher.Mono;
-import reactor.core.scheduler.Schedulers;
 
-import java.math.BigDecimal;
-import java.util.*;
-import java.util.concurrent.atomic.AtomicInteger;
-import java.util.concurrent.atomic.AtomicReference;
+import java.util.Map;
 
 /*
  * @author: ziin
@@ -46,12 +35,6 @@ import java.util.concurrent.atomic.AtomicReference;
 @Tag(name = "聊天", description = "聊天接口")
 public class ChatController {
 
-    // 最大消息长度限制
-    private static final int MAX_MESSAGE_LENGTH = 1000;
-
-    @Resource
-    private ChatClient client;
-
     @Resource
     private OpenAiEmbeddingModel embeddingModel;
 
@@ -59,174 +42,13 @@ public class ChatController {
     private QdrantVectorService qdrantVectorService;
 
     @Resource
-    private KeyboardCharacterService keyboardCharacterService;
-
-    @Resource
-    private KeyboardUserCallLogService callLogService;
+    private ChatService chatService;
 
 
     @PostMapping("/talk")
     @Operation(summary = "聊天润色接口", description = "聊天润色接口")
     public Flux<ServerSentEvent<ChatStreamMessage>> talk(@RequestBody ChatReq chatReq){
-        // 1. 参数校验
-        if (chatReq == null) {
-            log.error("聊天请求参数为空");
-            throw new BusinessException(ErrorCode.PARAMS_ERROR);
-        }
-
-        if (chatReq.getCharacterId() == null) {
-            log.error("键盘人设ID为空");
-            throw new BusinessException(ErrorCode.CHAT_CHARACTER_ID_EMPTY);
-        }
-
-        if (StrUtil.isBlank(chatReq.getMessage())) {
-            log.error("聊天消息为空");
-            throw new BusinessException(ErrorCode.CHAT_MESSAGE_EMPTY);
-        }
-
-        if (chatReq.getMessage().length() > MAX_MESSAGE_LENGTH) {
-            log.error("聊天消息过长，长度: {}", chatReq.getMessage().length());
-            throw new BusinessException(ErrorCode.CHAT_MESSAGE_TOO_LONG);
-        }
-
-        // 2. 验证键盘人设是否存在
-        KeyboardCharacter character = keyboardCharacterService.getById(chatReq.getCharacterId());
-        if (character == null) {
-            log.error("键盘人设不存在，ID: {}", chatReq.getCharacterId());
-            throw new BusinessException(ErrorCode.CHAT_CHARACTER_NOT_FOUND);
-        }
-
-        // 初始化调用日志
-        String requestId = IdUtil.fastSimpleUUID();
-        long startTime = System.currentTimeMillis();
-        AtomicReference<String> modelRef = new AtomicReference<>();
-        AtomicInteger inputTokens = new AtomicInteger(0);
-        AtomicInteger outputTokens = new AtomicInteger(0);
-        AtomicReference<String> errorCodeRef = new AtomicReference<>();
-
-        // 3. LLM 流式输出
-        Flux<ChatStreamMessage> llmFlux = client
-                .prompt(character.getPrompt())
-                .system("""
-                Format rules:
-                - Return EXACTLY 3 replies.
-                - Use "<SPLIT>" as the separator.
-                - reply1<SPLIT>reply2<SPLIT>reply3
-            """)
-                .user(chatReq.getMessage())
-                .options(OpenAiChatOptions.builder()
-                        .user(StpUtil.getLoginIdAsString())
-                        .build())
-                .stream()
-                .chatResponse()
-                .concatMap(response -> {
-                    // 提取 metadata
-                    if (response.getMetadata() != null) {
-                        var metadata = response.getMetadata();
-                        if (metadata.getModel() != null) {
-                            modelRef.set(metadata.getModel());
-                        }
-                        if (metadata.getUsage() != null) {
-                            var usage = metadata.getUsage();
-                            if (usage.getPromptTokens() != null) {
-                                inputTokens.set(usage.getPromptTokens());
-                            }
-                            if (usage.getCompletionTokens() != null) {
-                                outputTokens.set(usage.getCompletionTokens());
-                            }
-                        }
-
-
-                    }
-
-                    // 获取内容
-                    String content = response.getResult().getOutput().getText();
-                    if (content == null || content.isEmpty()) {
-                        return Flux.empty();
-                    }
-
-                    // 拆成单字符
-                    List<String> chars = content.codePoints()
-                            .mapToObj(cp -> new String(Character.toChars(cp)))
-                            .toList();
-
-                    // 按 3 个字符批量发送
-                    List<String> batched = new ArrayList<>();
-                    StringBuilder sb = new StringBuilder();
-                    for (String ch : chars) {
-                        sb.append(ch);
-                        if (sb.length() >= 3) {
-                            batched.add(sb.toString());
-                            sb.setLength(0);
-                        }
-                    }
-                    if (!sb.isEmpty()) {
-                        batched.add(sb.toString());
-                    }
-
-                    return Flux.fromIterable(batched)
-                            .map(s -> new ChatStreamMessage("llm_chunk", s));
-                })
-                .doOnError(error -> {
-                    log.error("LLM调用失败", error);
-                    errorCodeRef.set("LLM_ERROR");
-                })
-                .onErrorResume(error ->
-                    Flux.just(new ChatStreamMessage("error", "LLM服务暂时不可用，请稍后重试"))
-                );
-
-        // 4. 向量搜索Flux（一次性发送搜索结果）
-        Flux<ChatStreamMessage> searchFlux = Mono
-                .fromCallable(() -> qdrantVectorService.searchText(chatReq.getMessage()))
-                .subscribeOn(Schedulers.boundedElastic()) // 避免阻塞 event-loop
-                .map(list -> new ChatStreamMessage("search_result", list))
-                .doOnError(error -> log.error("向量搜索失败", error))
-                .onErrorResume(error ->
-                    Mono.just(new ChatStreamMessage("search_result", new ArrayList<>()))
-                )
-                .flux();
-
-        // 5. 结束标记
-        Flux<ChatStreamMessage> doneFlux =
-                Flux.just(new ChatStreamMessage("done", null));
-
-        // 6. 合并所有Flux
-        Flux<ChatStreamMessage> merged =
-                Flux.merge(llmFlux, searchFlux)
-                        .concatWith(doneFlux);
-        String tokenValue = StpUtil.getTokenValue();
-        // 7. SSE 包装并记录调用日志
-        return merged
-                .doFinally(signalType -> {
-                    // 异步保存调用日志
-                    Mono.fromRunnable(() -> {
-                        try {
-                            KeyboardUserCallLog callLog = new KeyboardUserCallLog();
-                            SaTokenContextMockUtil.setMockContext(()->{
-                                StpUtil.setTokenValueToStorage(tokenValue);
-                               callLog.setUserId(StpUtil.getLoginIdAsLong());
-                            });
-                            callLog.setRequestId(requestId);
-                            callLog.setFeature("chat_talk");
-                            callLog.setModel(modelRef.get());
-                            callLog.setInputTokens(inputTokens.get());
-                            callLog.setOutputTokens(outputTokens.get());
-                            callLog.setTotalTokens(inputTokens.get() + outputTokens.get());
-                            callLog.setSuccess(errorCodeRef.get() == null);
-                            callLog.setLatencyMs((int) (System.currentTimeMillis() - startTime));
-                            callLog.setErrorCode(errorCodeRef.get());
-                            callLog.setCreatedAt(new Date());
-                            callLogService.save(callLog);
-                        } catch (Exception e) {
-                            log.error("保存调用日志失败", e);
-                        }
-                    }).subscribeOn(Schedulers.boundedElastic()).subscribe();
-                })
-                .map(msg ->
-                        ServerSentEvent.builder(msg)
-                                .event(msg.getType())
-                                .build()
-                );
+        return chatService.talk(chatReq);
     }