fix: address SPEC-21 code quality issues

- Remove dead ContextVar code from token_tracking.py (was never set)
- Replace orphaned get_token_stats() with instance method get_stats()
- Add proper exports to workflows/__init__.py (SubIterationMiddleware, etc.)
- Clean up imports in huggingface.py (use package import)
- Add type: ignore for mypy list invariance issue

All 311 tests pass.

src/clients/huggingface.py

@@ -27,8 +27,7 @@ from agent_framework._types import FunctionCallContent, FunctionResultContent
 from agent_framework.observability import use_observability
 from huggingface_hub import InferenceClient
 
-from src.middleware.retry import RetryMiddleware
-from src.middleware.token_tracking import TokenTrackingMiddleware
+from src.middleware import RetryMiddleware, TokenTrackingMiddleware
 from src.utils.config import settings
 
 logger = structlog.get_logger()
@@ -53,13 +52,13 @@ class HuggingFaceChatClient(BaseChatClient):  # type: ignore[misc]
             api_key: HF_TOKEN (optional, defaults to env var).
             **kwargs: Additional arguments passed to BaseChatClient.
         """
-        # Create middleware instances
-        middleware: list[Any] = [
+        # Create middleware instances for retry and token tracking
+        middleware = [
             RetryMiddleware(max_attempts=3, min_wait=1.0, max_wait=10.0),
             TokenTrackingMiddleware(),
         ]
 
-        super().__init__(middleware=middleware, **kwargs)
+        super().__init__(middleware=middleware, **kwargs)  # type: ignore[arg-type]
         # FIX: Use 7B model to stay on HuggingFace native infrastructure (avoid Novita 500s)
         self.model_id = model_id or settings.huggingface_model or "Qwen/Qwen2.5-7B-Instruct"
         self.api_key = api_key or settings.hf_token

src/middleware/token_tracking.py

@@ -1,16 +1,12 @@
 """Token tracking middleware for monitoring API usage."""
 
 from collections.abc import Awaitable, Callable
-from contextvars import ContextVar
 
 import structlog
 from agent_framework._middleware import ChatContext, ChatMiddleware
 
 logger = structlog.get_logger()
 
-# ContextVar for per-request token tracking
-_request_tokens: ContextVar[dict[str, int]] = ContextVar("request_tokens")
-
 
 class TokenTrackingMiddleware(ChatMiddleware):
     """Tracks token usage across chat requests.
@@ -64,10 +60,14 @@ class TokenTrackingMiddleware(ChatMiddleware):
                 total_requests=self.request_count,
             )
 
-
-def get_token_stats() -> dict[str, int]:
-    """Get current request's token usage."""
-    try:
-        return _request_tokens.get().copy()
-    except LookupError:
-        return {"input": 0, "output": 0}
+    def get_stats(self) -> dict[str, int]:
+        """Get cumulative token usage statistics.
+
+        Returns:
+            Dictionary with total_input, total_output, and request_count.
+        """
+        return {
+            "total_input": self.total_input_tokens,
+            "total_output": self.total_output_tokens,
+            "request_count": self.request_count,
+        }

src/workflows/__init__.py

@@ -1 +1,13 @@
-"""Middleware components for orchestration."""
+"""Workflow components for orchestration.
+
+These are workflow patterns (e.g., team→judge loops), NOT interceptor middleware.
+For interceptor middleware, see src/middleware/.
+"""
+
+from src.workflows.sub_iteration import (
+    SubIterationJudge,
+    SubIterationMiddleware,
+    SubIterationTeam,
+)
+
+__all__ = ["SubIterationJudge", "SubIterationMiddleware", "SubIterationTeam"]