fix: resolve audited dispatch, kernel, and rollout bugs

- re-register the linear family with the operator dispatcher (ASTR_OPS / op_backend / resolve)
- fix bf16 gemv misaligned-address faults and element mispairing for offset weights
- reject misaligned bf16_swiglu inputs with a clear error and fall back in the backend gate
- make the rollout reuse decision, validation, and return atomic under one policy snapshot
- add the documented post-scoring rollout version check
- derive live+1 under the scheduler lock in optimizer_step via apply_weight_update(None, ...)
- reject rollout_max_policy_lag below rollout_interval - 1 at config time
- sync gemv stream-test inputs before switching streams; drop dead loader imports

tests/trainer/test_online_strategy.py

@@ -62,6 +62,9 @@ class _RecordingRunner:
 
     def apply_weight_update(self, policy_version, update):
         result = update()
+        if policy_version is None:
+            # Mirror the scheduler: None derives live+1 under the lock.
+            policy_version = self.policy_version + 1
         self.update_weights(policy_version)
         return result