feat(safety): enhance rate limiting and cost control with alert deduplication and usage tracking

- Added `record_action` in `RateLimiter` for precise tracking of slot consumption post-validation.
- Introduced deduplication mechanism for warning alerts in `CostController` to prevent spamming.
- Refactored `CostController`'s session and daily budget alert handling for improved clarity.
- Implemented test suites for `CostController` and `SafetyGuardian` to validate changes.
- Expanded integration testing to cover deduplication, validation, and loop detection edge cases.

backend/app/services/safety/limits/limiter.py

@@ -223,7 +223,10 @@ class RateLimiter:
         action: ActionRequest,
     ) -> tuple[bool, list[RateLimitStatus]]:
         """
-        Check all applicable rate limits for an action.
+        Check all applicable rate limits for an action WITHOUT consuming slots.
+
+        Use this during validation to check if action would be allowed.
+        Call record_action() after successful execution to consume slots.
 
         Args:
             action: The action to check
@@ -235,28 +238,53 @@ class RateLimiter:
         statuses: list[RateLimitStatus] = []
         allowed = True
 
-        # Check general actions limit
-        actions_allowed, actions_status = await self.acquire("actions", agent_id)
+        # Check general actions limit (read-only)
+        actions_status = await self.check("actions", agent_id)
         statuses.append(actions_status)
-        if not actions_allowed:
+        if actions_status.is_limited:
             allowed = False
 
         # Check LLM-specific limit for LLM calls
         if action.action_type.value == "llm_call":
-            llm_allowed, llm_status = await self.acquire("llm_calls", agent_id)
+            llm_status = await self.check("llm_calls", agent_id)
             statuses.append(llm_status)
-            if not llm_allowed:
+            if llm_status.is_limited:
                 allowed = False
 
         # Check file ops limit for file operations
         if action.action_type.value in {"file_read", "file_write", "file_delete"}:
-            file_allowed, file_status = await self.acquire("file_ops", agent_id)
+            file_status = await self.check("file_ops", agent_id)
             statuses.append(file_status)
-            if not file_allowed:
+            if file_status.is_limited:
                 allowed = False
 
         return allowed, statuses
 
+    async def record_action(
+        self,
+        action: ActionRequest,
+    ) -> None:
+        """
+        Record an action by consuming rate limit slots.
+
+        Call this AFTER successful execution to properly count the action.
+
+        Args:
+            action: The executed action
+        """
+        agent_id = action.metadata.agent_id
+
+        # Consume general actions slot
+        await self.acquire("actions", agent_id)
+
+        # Consume LLM-specific slot for LLM calls
+        if action.action_type.value == "llm_call":
+            await self.acquire("llm_calls", agent_id)
+
+        # Consume file ops slot for file operations
+        if action.action_type.value in {"file_read", "file_write", "file_delete"}:
+            await self.acquire("file_ops", agent_id)
+
     async def require(
         self,
         limit_name: str,