Use tx-pooler for pg connection and add index

[roeierez]

Improving performance by:

1. Add missing index on records(user_id, revision)
2. Disable prepare statement mode in postgress (so we can use stateless tx pooloer which scales a lot better)
3. To be able to seamlessly rollback if needed added two DB env variables. the migration db still need to be session pooler on some platforms as it uses advisory locks.
4. Allow configure max connections on the pool (right now we are using default which is a bit low).

[dangeross]

Perhaps we're the only users of data-sync, but do you think it maybe safer to fail fast if (the old) DATABASE_URL is set and DATABASE_RUNTIME_URL not set?

[roeierez]

That will change the semantic of which data source to use. I honestly think it is ok and it will allow us rollback fast if needed.

[dangeross]

My concern is more for other users of it, on update if they don't change their env config they'll switch into sqlite storage

[dangeross]

Should we log a warning that the DATABASE_RUNTIME_URL will be used?

[roeierez]

Added log.

[dangeross]

Claude suggests using CREATE INDEX CONCURRENTLY as it otherwise takes exclusive access, blocking other uses of records until complete. Depends how large the table is I guess

[roeierez]

Makes sense. Added

[JssDWt]

What is the reason to change this index? I'm asking because the query performance advisor doesn't flag this as a suggestion. It does suggest these for about 5% performance gain.

CREATE INDEX ON public.records USING btree (user_id);

and

CREATE INDEX ON public.user_revisions USING btree (user_id);