fix: address CodeRabbit review feedback for SPEC-21

- token_tracking.py: Handle both dict and Pydantic object access for usage
- retry.py: Add jitter (±25%) to exponential backoff to prevent thundering herd
- test_retry.py: Add pytestmark = pytest.mark.unit per coding guidelines
- test_token_tracking.py: Add pytestmark = pytest.mark.unit per coding guidelines

src/middleware/retry.py

@@ -1,6 +1,7 @@
 """Retry middleware for chat clients with exponential backoff."""
 
 import asyncio
+import random
 from collections.abc import Awaitable, Callable
 
 import structlog
@@ -52,9 +53,12 @@ class RetryMiddleware(ChatMiddleware):
         return False
 
     def _calculate_wait(self, attempt: int) -> float:
-        """Calculate wait time with exponential backoff."""
+        """Calculate wait time with exponential backoff and jitter."""
         wait = self.min_wait * (2**attempt)
-        return float(min(wait, self.max_wait))
+        wait = min(wait, self.max_wait)
+        # Add jitter (±25%) to avoid thundering herd
+        jitter = wait * 0.25 * (2 * random.random() - 1)
+        return float(max(self.min_wait, wait + jitter))
 
     async def process(
         self, context: ChatContext, next: Callable[[ChatContext], Awaitable[None]]

src/middleware/token_tracking.py

@@ -44,8 +44,19 @@ class TokenTrackingMiddleware(ChatMiddleware):
                 usage = msg.metadata.get("usage")
 
         if usage:
-            input_tokens = usage.get("input_tokens", 0) or usage.get("prompt_tokens", 0)
-            output_tokens = usage.get("output_tokens", 0) or usage.get("completion_tokens", 0)
+            # Handle both dict-like and object attribute access
+            if hasattr(usage, "get"):
+                # Dict-like access
+                input_tokens = usage.get("input_tokens", 0) or usage.get("prompt_tokens", 0)
+                output_tokens = usage.get("output_tokens", 0) or usage.get("completion_tokens", 0)
+            else:
+                # Object attribute access (Pydantic models, etc.)
+                input_tokens = getattr(usage, "input_tokens", 0) or getattr(
+                    usage, "prompt_tokens", 0
+                )
+                output_tokens = getattr(usage, "output_tokens", 0) or getattr(
+                    usage, "completion_tokens", 0
+                )
 
             self.total_input_tokens += input_tokens
             self.total_output_tokens += output_tokens

tests/unit/middleware/test_retry.py

@@ -4,6 +4,8 @@ import pytest
 
 from src.middleware.retry import RetryMiddleware
 
+pytestmark = pytest.mark.unit
+
 
 @pytest.mark.asyncio
 async def test_retry_middleware_succeeds_first_try():

tests/unit/middleware/test_token_tracking.py

@@ -4,6 +4,8 @@ import pytest
 
 from src.middleware.token_tracking import TokenTrackingMiddleware
 
+pytestmark = pytest.mark.unit
+
 
 @pytest.mark.asyncio
 async def test_token_tracking_middleware_counts_tokens():