fix: Metastore sync tables with a single transaction

[baumandm]

We were facing some database perf issues, so I looked into it and the SQL query with the most waits was INSERT INTO 'data_table_warnings':

We recently updated our metastore loader code to automatically add warnings to certain tables during the sync. That explains why we're seeing it now, but it didn't seem right so I looked into it.

While most of the metastore sync executes as a single transaction per table, I realized data table warnings are being committed immediately. I enabled database logging in SQLAlchemy and ran through the metastore sync, which helped identify a few other areas where separate transactions were being committed.

This PR fixes every case of a separate transaction in the base_metastore_loader's _create_table_table function that I could detect.

1. Where necessary, I added a commit arg to functions that were missing it (and passing commit=False during the sync)
2. Functions that have a commit arg should always pass commit=False to all sub-query functions, since they (optionally) commit at the end of the function

With these changes, each table creates a single transaction that is committed (or rolled back) at the completion of the _create_table_table() function.

We deployed this change to production late on July 30th, and the load dropped significantly; all the row_lock_waits are gone:

Moreover, the metastore loader task runs 100% faster now:

Note: to enable database logging, add echo=True to the create_engine() call in db.py. This let me know exactly when transactions were being committed.

[czgu]

Nice work!