Migrate psycopg2 -> psycopg 3

[seroperson]

Closes #902

Summary

1. Replace psycopg2-binary with psycopg[binary]>=3.2
2. Drop execute_values (unsupported in v3) and switch each bulk upsert to cursor.executemany, with the SQL templates expanded from VALUES %s to explicit VALUES (%s, %s, ...) row tuples
3. Wrap all per-miner writes inside db_connection.pipeline() to batch queries into one round-trip while preserving the existing transaction boundary. From documentation:

"The pipeline mode allows PostgreSQL client applications to send a query without having to read the result of the previously sent query... multiple queries/results can be sent/received in a single network roundtrip."

4. Set connection.prepare_threshold = 0 so our queries are auto-prepared, instead of old behavior, when it was undefined. From documentation:

"A query is prepared automatically after it is executed more than prepare_threshold times on a connection..."

5. No schema change, no public API change on Repository or DatabaseStorage
6. Use dbname arg instead of database, like in canonical usage example: https://www.psycopg.org/psycopg3/docs/basic/usage.html#main-objects-in-psycopg-3; Keyword list: https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-PARAMKEYWORDS

Test plan

I haven't added new tests as currently there are no any tests which cover the real database connection, either any migrations to spin-up everything locally. I did some synthetic manual testing and everything seems good.

I can try to implement necessary code here to test it against a real connection and do real tests if you want.

[anderdc]

1. Fix conflicts.
2. Pin psycopg[binary] to a specific version (e.g. ==3.3.3) — rest of pyproject.toml uses exact pins.