fix(planner): guard apply_payload and reset_to_defaults with a lock

Concurrent PUT requests could observe a half-updated planner between
the four sequential global assignments. Added _planner_lock so the
rebind is atomic; same lock wraps reset_to_defaults.

decnet/realism/planner.py

@@ -20,6 +20,7 @@ persona outside its window is never considered.
 from __future__ import annotations
 
 import secrets
+import threading
 from datetime import datetime
 from typing import Any, Optional, Sequence
 
@@ -74,6 +75,7 @@ _USER_CLASS_WEIGHTS: tuple[tuple[ContentClass, int], ...] = _DEFAULT_USER_CLASS_
 _SYSTEM_CLASS_WEIGHTS: tuple[tuple[ContentClass, int], ...] = _DEFAULT_SYSTEM_CLASS_WEIGHTS
 _CANARY_CLASS_WEIGHTS: tuple[tuple[ContentClass, int], ...] = _DEFAULT_CANARY_CLASS_WEIGHTS
 _CANARY_PROBABILITY: float = _DEFAULT_CANARY_PROBABILITY
+_planner_lock = threading.Lock()
 
 
 def _serialize_weights(
@@ -185,6 +187,7 @@ def apply_payload(payload: dict[str, Any]) -> None:
             raise ValueError("canary_probability must be in [0.0, 1.0]")
         new_prob = float(prob)
 
+    with _planner_lock:
         _USER_CLASS_WEIGHTS = new_user
         _SYSTEM_CLASS_WEIGHTS = new_system
         _CANARY_CLASS_WEIGHTS = new_canary
@@ -195,6 +198,7 @@ def reset_to_defaults() -> None:
     """Restore hardcoded defaults. Used by tests and the API reset path."""
     global _USER_CLASS_WEIGHTS, _SYSTEM_CLASS_WEIGHTS
     global _CANARY_CLASS_WEIGHTS, _CANARY_PROBABILITY
+    with _planner_lock:
         _USER_CLASS_WEIGHTS = _DEFAULT_USER_CLASS_WEIGHTS
         _SYSTEM_CLASS_WEIGHTS = _DEFAULT_SYSTEM_CLASS_WEIGHTS
         _CANARY_CLASS_WEIGHTS = _DEFAULT_CANARY_CLASS_WEIGHTS