Summary

Adds pgpm tune — a CLI command that disables PostgreSQL durability for throwaway databases (CI service containers, local dev, test containers). Motivated by CI flakiness in constructive-db (constructive-io/constructive-db#1925): WAL checkpoints stalling on runner disk I/O were pushing DDL-heavy provisioning tests past their jest timeouts. With this command, that repo's workflow psql one-liner can become pgpm tune --yes.

Applies via ALTER SYSTEM + pg_reload_conf() (all SIGHUP-reloadable — no restart):

fsync = off
synchronous_commit = off
full_page_writes = off
checkpoint_timeout = 30min
max_wal_size = 8GB

Usage:

pgpm tune --yes          # non-interactive (CI)
pgpm tune --reset --yes  # ALTER SYSTEM RESET back to server defaults

Follows the admin-users bootstrap pattern (prompter confirm gated by --yes, getPgPool from pg-cache, Logger); registered in createPgpmCommandMap with the pg-teardown wrapper. Verified against a live postgres-plus:18 container: settings apply and reset correctly.

Note: pnpm lint currently fails repo-wide (ESLint 9 installed but config is .eslintrc.json) — pre-existing, unrelated to this change.

Link to Devin session: https://app.devin.ai/sessions/7943c7526d254aa4828be110fbb3f272
Requested by: @pyramation