fix(awooop): cache heavy operator summaries

apps/api/src/services/platform_operator_service.py

@@ -9,6 +9,7 @@ ADR-106（AwoooP Agent Platform）
 from __future__ import annotations
 
 import asyncio
+import os
 import re
 import time
 import uuid
@@ -59,6 +60,10 @@ from src.services.ollama_failover_manager import (
 )
 from src.services.ollama_health_monitor import HealthReport, HealthStatus
 from src.services.operator_outcome import build_operator_outcome
+from src.services.operator_summary_cache import (
+    get_cached_operator_summary,
+    store_operator_summary,
+)
 from src.services.run_state_machine import transition
 
 logger = structlog.get_logger(__name__)
@@ -74,6 +79,9 @@ _MAX_STEP_SUMMARY_CHARS = 128
 _AI_ROUTE_STATUS_SELECT_TIMEOUT_SECONDS = 12.0
 _AI_ROUTE_STATUS_CONNECTIVITY_TIMEOUT_SECONDS = 2.5
 _REMEDIATION_HISTORY_LIMIT = 20
+_CALLBACK_REPLY_CACHE_TTL_SECONDS = int(
+    os.getenv("AWOOOP_CALLBACK_REPLY_CACHE_TTL_SECONDS", "20")
+)
 _INCIDENT_ID_RE = re.compile(r"\bINC-\d{8}-[A-Z0-9]{4,}\b")
 _REMEDIATION_STATUS_FILTERS = {
     "mcp_observed",
@@ -304,11 +312,13 @@ async def list_callback_replies(
     incident_id: str | None,
     page: int,
     per_page: int,
+    refresh: bool = False,
 ) -> dict[str, Any]:
     """列出 Telegram detail/history callback reply evidence，不改 runtime 狀態。"""
     _validate_callback_reply_status_filter(callback_reply_status)
     callback_action = _validate_callback_reply_action_filter(action)
     _validate_incident_id_filter(incident_id)
+    normalized_project_id = project_id or "awoooi"
 
     if callback_reply_status == "no_callback":
         return {
@@ -318,6 +328,33 @@ async def list_callback_replies(
             "per_page": per_page,
         }
 
+    cache_key = {
+        "project_id": project_id or "__all__",
+        "callback_reply_status": callback_reply_status or "",
+        "action": callback_action or "",
+        "incident_id": incident_id or "",
+        "page": page,
+        "per_page": per_page,
+    }
+    if not refresh:
+        cached_response = get_cached_operator_summary(
+            "callback_replies",
+            cache_key,
+            ttl_seconds=_CALLBACK_REPLY_CACHE_TTL_SECONDS,
+        )
+        if cached_response is not None:
+            logger.info(
+                "operator_callback_replies_cache_hit",
+                project_id=normalized_project_id,
+                callback_reply_status=callback_reply_status,
+                action=callback_action,
+                incident_id=incident_id,
+                page=page,
+                per_page=per_page,
+                ttl_seconds=_CALLBACK_REPLY_CACHE_TTL_SECONDS,
+            )
+            return cached_response
+
     where_clauses = [
         "m.source_envelope ? 'callback_reply'",
     ]
@@ -399,14 +436,14 @@ async def list_callback_replies(
         LIMIT :limit OFFSET :offset
     """)
 
-    async with get_db_context(project_id or "awoooi") as db:
+    async with get_db_context(normalized_project_id) as db:
         count_result = await db.execute(count_sql, params)
         total = count_result.scalar_one()
         rows_result = await db.execute(list_sql, params)
         rows = list(rows_result.mappings().all())
         audit_summary = await _fetch_callback_reply_audit_summary(
             db,
-            project_id=project_id or "awoooi",
+            project_id=normalized_project_id,
         )
 
     items = [_callback_reply_event_item(row) for row in rows]
@@ -459,13 +496,31 @@ async def list_callback_replies(
             km_completion_summary_cache[summary_cache_key] = km_summary
         item["km_stale_completion_summary"] = km_summary
 
-    return {
+    response = {
         "items": items,
         "total": total,
         "page": page,
         "per_page": per_page,
         "summary": audit_summary,
     }
+    logger.info(
+        "operator_callback_replies_fetched",
+        project_id=normalized_project_id,
+        callback_reply_status=callback_reply_status,
+        action=callback_action,
+        incident_id=incident_id,
+        page=page,
+        per_page=per_page,
+        total=total,
+        cache_status="miss",
+        cache_ttl_seconds=_CALLBACK_REPLY_CACHE_TTL_SECONDS,
+    )
+    return store_operator_summary(
+        "callback_replies",
+        cache_key,
+        response,
+        ttl_seconds=_CALLBACK_REPLY_CACHE_TTL_SECONDS,
+    )
 
 
 async def _fetch_callback_reply_audit_summary(