fix(db): create base schema before running migrations

Migrations ran before `_Base.metadata.create_all()`, so migration 3->4
  (`ALTER TABLE user_mappings ADD COLUMN avatar_url`) failed on a fresh or
  partially-initialized database where the ORM-model tables did not yet
  exist — no migration creates the core tables (user_mappings,
  message_mappings, user_bindings, binding_codes); they are defined only
  by the ORM models, while migrations 0->3 touch only the forward_* tables.

  Move `create_all` ahead of `_run_migrations` in `MessageDB.__init__` and
  drop the migration call from `_create_engine_from_config`. Init is now
  self-healing: create_all builds the current schema (recreating any missing
  tables), then idempotent migrations apply incremental ALTERs.

  Fixes "Startup aborted: database initialization failed".

Author: NoClassFoundError

services/db.py

@@ -119,7 +119,13 @@ def __init__(self, engine: Engine | None = None):
             engine: Optional SQLAlchemy engine. If not provided, will be created from config.
         """
         self._engine = engine or self._create_engine_from_config()
+        # Create the base ORM schema BEFORE running incremental migrations.
+        # Tables like `user_mappings` are defined only by the ORM models, and
+        # migrations (e.g. 3->4 `ALTER TABLE user_mappings ...`) assume they
+        # already exist. Running migrations first would fail on a fresh or
+        # partially-initialized database.
         _Base.metadata.create_all(self._engine)
+        self._run_migrations(self._engine)
 
     @staticmethod
     def _create_engine_from_config() -> Engine:
@@ -180,8 +186,6 @@ def _create_engine_from_config() -> Engine:
         logger.info(f"Initializing database engine: {url.split('://')[0]}")
         engine = create_engine(url, **engine_kwargs)
 
-        MessageDB._run_migrations(engine)
-
         return engine
 
     @staticmethod