feat: 实现日历提醒完整功能（操作执行、通知服务重构、归档）

- 新增 ReminderActionExecutor 处理取消/稍后提醒操作
- 新增 ReminderOutboxStore 本地存储待处理操作
- 重构 LocalNotificationService 支持聚合提醒和交互操作
- 新增 event_color_resolver 工具类统一颜色解析
- 新增 CalendarService.archiveEvent 归档方法
- 增强 ModelTracking 支持缓存命中、推理token和成本追踪
- 添加 qwen3.5-35b-a3b 模型配置
- 更新 AndroidManifest 全屏intent权限
- 补充相关单元测试和文档

backend/src/services/litellm/service.py

@@ -85,9 +85,15 @@ class LiteLLMService:
                 selected_tier = tier
                 break
 
+        cached_token_rate = (
+            selected_tier.cache_hit_cost_per_token
+            if selected_tier.cache_hit_cost_per_token > 0
+            else selected_tier.input_cost_per_token
+        )
+
         return float(
             uncached_prompt_tokens * selected_tier.input_cost_per_token
-            + normalized_cached_tokens * selected_tier.cache_hit_cost_per_token
+            + normalized_cached_tokens * cached_token_rate
             + normalized_completion_tokens * selected_tier.output_cost_per_token
         )
 
@@ -95,23 +101,86 @@ class LiteLLMService:
         self,
         *,
         model: str,
-        usage_summary: dict[str, int] | None,
+        usage_summary: dict[str, Any] | None,
     ) -> dict[str, Any]:
         summary = usage_summary or {}
         input_tokens = max(int(summary.get("input_tokens", 0) or 0), 0)
         output_tokens = max(int(summary.get("output_tokens", 0) or 0), 0)
+        total_tokens = max(
+            int(summary.get("total_tokens", input_tokens + output_tokens) or 0), 0
+        )
         latency_ms = max(int(summary.get("latency_ms", 0) or 0), 0)
         cached_prompt_tokens = max(int(summary.get("cached_prompt_tokens", 0) or 0), 0)
-        cost = self.calculate_cost(
-            model=model,
-            prompt_tokens=input_tokens,
-            completion_tokens=output_tokens,
-            cached_prompt_tokens=cached_prompt_tokens,
+        prompt_cache_hit_tokens = max(
+            int(summary.get("prompt_cache_hit_tokens", cached_prompt_tokens) or 0), 0
         )
+        prompt_cache_miss_tokens = max(
+            int(
+                summary.get(
+                    "prompt_cache_miss_tokens",
+                    max(input_tokens - prompt_cache_hit_tokens, 0),
+                )
+                or 0
+            ),
+            0,
+        )
+        reasoning_tokens = max(int(summary.get("reasoning_tokens", 0) or 0), 0)
+        direct_cost_raw = summary.get("direct_cost")
+        direct_cost_observed = bool(int(summary.get("direct_cost_observed", 0) or 0))
+        direct_cost_complete = bool(int(summary.get("direct_cost_complete", 0) or 0))
+        model_call_records = max(int(summary.get("model_call_records", 0) or 0), 0)
+        usage_records = max(int(summary.get("usage_records", 0) or 0), 0)
+        usage_complete = model_call_records == 0 or model_call_records == usage_records
+        direct_cost = self._coerce_non_negative_float(direct_cost_raw)
+
+        if (
+            usage_complete
+            and direct_cost_observed
+            and direct_cost_complete
+            and direct_cost is not None
+        ):
+            cost = direct_cost
+            cost_source = "provider"
+        else:
+            cost = self.calculate_cost(
+                model=model,
+                prompt_tokens=input_tokens,
+                completion_tokens=output_tokens,
+                cached_prompt_tokens=cached_prompt_tokens,
+            )
+            cost_source = (
+                "incomplete_usage_fallback"
+                if not usage_complete
+                else (
+                    "catalog_fallback_incomplete_provider_cost"
+                    if direct_cost_observed and not direct_cost_complete
+                    else "catalog_fallback"
+                )
+            )
+
         return {
             "model": model,
             "inputTokens": input_tokens,
             "outputTokens": output_tokens,
+            "totalTokens": total_tokens,
+            "cachedPromptTokens": cached_prompt_tokens,
+            "promptCacheHitTokens": prompt_cache_hit_tokens,
+            "promptCacheMissTokens": prompt_cache_miss_tokens,
+            "reasoningTokens": reasoning_tokens,
             "cost": cost,
+            "costSource": cost_source,
+            "usageComplete": usage_complete,
             "latencyMs": latency_ms,
         }
+
+    @staticmethod
+    def _coerce_non_negative_float(value: Any) -> float | None:
+        if value is None:
+            return None
+        try:
+            parsed = float(value)
+        except (TypeError, ValueError):
+            return None
+        if parsed < 0:
+            return None
+        return parsed