### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

The `POST /global/spend/refresh` endpoint (`global_spend_refresh()`) instantiates a dedicated `PrismaClient` for the `REFRESH MATERIALIZED VIEW` step, calls `.connect()` on it, runs the query, and then returns **without ever calling `.disconnect()`**. Every call to this endpoint leaks a Prisma engine + its DB connection. Repeated refreshes (the Admin UI Usage tab, or any polling/monitoring that hits this endpoint) steadily accumulate idle Postgres connections until the DB reaches `max_connections` and the whole proxy starts failing with connection errors.

**Expected:** the endpoint adds zero net DB connections after it returns, no matter how many times it is called — the dedicated client must always be released.

**Actual:** each successful call opens a new Prisma engine/connection that is never released; the `idle` count in `pg_stat_activity` grows by one per call and never decreases, until `FATAL: sorry, too many clients already`.

### Steps to Reproduce

Reproducible on a stock proxy + Postgres, no special infra:

1. Start Postgres with a l…