fix(persistence): reclaim oversized SQLite WAL files

- Require successful WAL truncation before starting application producers
- Preserve completed migrations and retry checkpoint failures on restart
- Set a 64 MiB WAL retention limit on every SQLite connection
- Preserve default automatic checkpoint behavior and reader snapshots
- Cover startup failures, connection recreation, and WAL reclamation

src/langbot/pkg/persistence/sqlite_startup.py

@@ -0,0 +1,27 @@
+"""Finish SQLite WAL maintenance before application producers start."""
+
+STARTUP_BUSY_TIMEOUT_MS = 60_000
+
+
+async def checkpoint_after_migration(engine):
+    """Require a complete WAL checkpoint, including on a retry at schema head."""
+    if engine.dialect.name != 'sqlite':
+        return
+    async with engine.connect() as connection:
+        connection = await connection.execution_options(isolation_level='AUTOCOMMIT')
+        if (await connection.exec_driver_sql('PRAGMA journal_mode')).scalar_one().lower() != 'wal':
+            return
+        timeout = (await connection.exec_driver_sql('PRAGMA busy_timeout')).scalar_one()
+        try:
+            await connection.exec_driver_sql(f'PRAGMA busy_timeout={STARTUP_BUSY_TIMEOUT_MS}')
+            busy, frames, checkpointed = (
+                await connection.exec_driver_sql('PRAGMA main.wal_checkpoint(TRUNCATE)')
+            ).one()
+            if (busy, frames, checkpointed) != (0, 0, 0):
+                raise RuntimeError(
+                    'SQLite startup WAL checkpoint did not complete: '
+                    f'busy={busy}, frames={frames}, checkpointed={checkpointed}. '
+                    'Stop other database users and restart LangBot to retry.'
+                )
+        finally:
+            await connection.exec_driver_sql(f'PRAGMA busy_timeout={timeout}')