refactor: harden inference cache state and attention dispatch

- split KVCache into phase-specific PrefillKVCache/DecodeKVCache types selected by start_pos
- unify steady-state detection in TaskCacheManager
- guard decode steady-state reuse with the cached task signature so recycled req slots cannot replay a prior generation's tokens and positions
- collapse attention backend fwd_decode/fwd_prefill into a single subclass-owned forward with a shared _check_fwd guard
- fix thread-safety gap in weight update and validate prefill inputs before KV allocation
- centralize magic constants in InferenceConfig and align docs with behavior

astrai/inference/scheduler.py

@@ -150,6 +150,7 @@ class InferenceScheduler:
             )
 
     def _ensure_weight_update_ready(self) -> None:
+        """Check weight update preconditions. Must be called under _weight_lock."""
         if self._loop_thread is not None and self._loop_thread.is_alive():
             raise RuntimeError("Stop the scheduler before updating model weights")
         if self._task_mgr.get_active_tasks() or self._task_mgr.get_waiting_tasks():