fix(deepseek): pass requestId context to cache and widen sanitizer regex

- chatCore.ts: pass {requestId:skillRequestId,messageIndex:0} to cacheReasoningFromAssistantMessage
- responseSanitizer.ts: widen isDeepSeekV4Model regex to match all deepseek-v4 variants

open-sse/handlers/chatCore.ts

@@ -4134,7 +4134,7 @@ export async function handleChatCore({
     try {
       const firstChoice = translatedResponse?.choices?.[0];
       const msg = firstChoice?.message;
-      cacheReasoningFromAssistantMessage(msg, provider, model);
+      cacheReasoningFromAssistantMessage(msg, provider, model, { requestId: skillRequestId, messageIndex: 0 });
     } catch {
       // Cache capture is non-critical — never block the response
     }
@@ -4421,7 +4421,7 @@ export async function handleChatCore({
         const body = streamResponseBody as Record<string, unknown>;
         const choices = body.choices as { message?: Record<string, unknown> }[] | undefined;
         const msg = choices?.[0]?.message;
-        cacheReasoningFromAssistantMessage(msg, provider, model);
+        cacheReasoningFromAssistantMessage(msg, provider, model, { requestId: skillRequestId, messageIndex: 0 });
       } catch {
         // Cache capture is non-critical — never block the stream
       }

open-sse/handlers/responseSanitizer.ts

@@ -28,7 +28,7 @@ const ALLOWED_RESPONSES_USAGE_FIELDS = new Set([
 type JsonRecord = Record<string, unknown>;
 
 function isDeepSeekV4Model(model: unknown): boolean {
-  return typeof model === "string" && /^deepseek-v4(?:$|[-_/])/i.test(model);
+  return typeof model === "string" && /deepseek[-/]v4/i.test(model);
 }
 
 function toRecord(value: unknown): JsonRecord | null {