fix(usage): enforce spend caps for room turns

[casquijo]

What changed

• Capture immutable usage attribution when each room member is dispatched and record it when the provider turn completes.
• Recheck the workspace spend cap immediately before room and calendar provider dispatches.
• Preserve attribution across queued messages and identify routine-driven room goals correctly.
• Treat capped room goals as blocked instead of retrying them as provider failures.
• Add regression coverage for rooms, goals, queued users, and calendar calls.

Why

Direct turns were accounted for and capped, but provider-backed room and calendar turns could proceed without equivalent ledger entries or dispatch-time enforcement.

This ensures settled room usage is visible before the next member dispatch and that all affected paths observe the same monthly cap.

How it was verified

• pnpm typecheck
• pnpm lint
• Targeted spend-cap tests: 6 passed
• Affected and review suites: 147 passed, 1 todo
• pnpm test
  • Vitest: 461 files passed, 2 skipped; 5,731 tests passed, 40 skipped, 1 todo
  • Broker tests: 8 passed
  • Electron tests: 164 passed
  • Packaged-server smoke passed

Screenshots (UI changes)

Not applicable — no UI changes.

Checklist

• pnpm typecheck and pnpm test pass locally
• Server behavior changes come with tests
• No dist-server/ edits
• No macOS-only code or shell command construction added
• No secrets in logs, responses, events, or argv

[vercel]

@casquijo is attempting to deploy a commit to the SupaMaus Team on Vercel.

A member of the Team first needs to authorize it.

[milind-soni]

Triage update: #1135 is now merged, so the room spend-cap guard, room ledger entries, blocked-goal outcome, and live budget fields already exist on main. This PR is not wholly redundant: capturing the initiating user on each operation/queued item still matters. Current main reads turnTriggers at settlement, which a later queued sender can overwrite. Please rebase this as the narrower immutable-attribution follow-up, preserving the queued-user, calendar-owner, and routine-goal regressions, without adding a second spend/ledger path. Holding the current overlapping version rather than closing it as a duplicate.