Fix StateManager deadlock with explicit BEGIN IMMEDIATE + retry

with conn: context manager in Python sqlite3 doesn't respect busy_timeout
for the implicit BEGIN; under concurrent subprocess writes the transaction
would fail immediately. Replace with explicit BEGIN IMMEDIATE + retry loop
(6 attempts, 0.5s back-off) to handle WAL write contention.

core/state.py

@@ -250,12 +250,29 @@ class StateManager:
             return
 
         conn = self._get_write_conn()
+        for attempt in range(6):
             try:
-            with conn:
+                conn.execute("BEGIN IMMEDIATE")
                 for sql, params in batch:
                     conn.execute(sql, params)
+                conn.execute("COMMIT")
+                return
+            except sqlite3.OperationalError as e:
+                try:
+                    conn.execute("ROLLBACK")
                 except Exception:
+                    pass
+                if attempt < 5:
+                    time.sleep(0.5 * (attempt + 1))
+                else:
+                    logger.error("StateManager flush failed after 6 attempts (%d ops): %s", len(batch), e)
+            except Exception:
+                try:
+                    conn.execute("ROLLBACK")
+                except Exception:
+                    pass
                 logger.exception("StateManager flush failed (%d ops)", len(batch))
+                return
 
     def _writer_loop(self) -> None:
         """Background loop: flush write queue every FLUSH_INTERVAL seconds."""