EXE-1422: make pg setup more resilient

[jacobheric]

the old teardown executed all cleanup sql as a single statement. if any step failed (e.g. the replication slot was active, or a role had already been removed), the entire teardown would fail and leave resources behind. this made it impossible for users to remove and re-add their CDC integration. They'd then get errors like: replication slot "inngest_cdc" is active for PID <N (SQLSTATE 55006)" and get stuck.

Changes

• execute each teardown step individually so that a failure in one step (e.g. role already removed) doesn't block the rest of the cleanup.
• handle active replication slots by checking pg_replication_slots for an active PID, calling pg_terminate_backend(), and retrying the slot drop after a brief delay if needed.
• add defer conn.Close(ctx) to avoid leaking the teardown connection.
• best-effort cleanup for non-critical resources — privilege revocations and user/publication drops log warnings but don't fail the overall teardown if the resource is already gone.

NOTE: I vendored this into monorepo locally and tested it e2e against various failures scenarios with neon and supabase.

[linear]

EXE-1422 neon integration broken