fix: pool config+health (#596), SEP-24 contracts (#603), async PDF queue (#595), SOS Maestro E2E (#607)

[Creed1759]

Summary

Closes DogStark#596, DogStark#603, DogStark#595, DogStark#607 — four independent issues solved in one pass.

---

DogStark#596 — PostgreSQL connection pool configuration & monitoring

What: The pool was created with default/misconfigured settings and the /health endpoint had no pool visibility.

How:

• Replaced old env vars (DB_POOL_SIZE, DB_IDLE_TIMEOUT) with DB_POOL_MAX (default 20), DB_IDLE_TIMEOUT_MS (default 10000), DB_CONNECTION_TIMEOUT_MS (default 3000)
• Added getPoolStats() returning { total, idle, waiting }
• GET /api/health now returns pool stats alongside the existing response
• A console.warn fires on every new connection when waiting > 5
• New integration test in backend/tests/integration/poolExhaustion.test.ts verifies requests fail fast (not hang) after connectionTimeoutMillis

---

DogStark#603 — SEP-24 anchor contract tests

What: Anchor API changes could silently break the deposit/withdrawal flow.

How:

• Added backend/services/__tests__/stellarAnchorService.contract.test.ts
• Lightweight consumer-driven contracts using an inline mock provider (no extra Pact dependency)
• Contracts cover: stellar.toml shape, /deposit/interactive request/response, /withdraw/interactive with amount, /transaction poll (pending & complete states), terminal-status caching (no re-poll after completion), and the SEP-24 /info response shape
• If the anchor changes its API shape, the relevant assertion fails before the real integration breaks

---

DogStark#595 — Async PDF generation via Redis job queue

What: Synchronous PDF generation in the HTTP request timed out for large reports.

How:

• Rewrote backend/server/routes/reports.ts with three endpoints:
  • POST /api/reports/pets/:petId/health → returns 202 { jobId } immediately
  • GET /api/reports/:jobId/status → returns { status, recordCount, filename, error }
  • GET /api/reports/:jobId/download → streams PDF once complete
• PDF generation runs in setImmediate (background, non-blocking)
• Job state stored in Redis with 1-hour TTL using the existing getRedisClient()
• Graceful synchronous fallback if Redis is unavailable
• Failed jobs return a typed error that allows client-side retry
• DocumentVaultScreen.tsx updated: new "Generate Health Report" button enqueues the job, polls /status every 2 s with an activity indicator, and shows a "Download PDF" button when status === 'complete'

---

DogStark#607 — Maestro E2E flow for Emergency SOS

What: No Maestro coverage for the highest-stakes feature in the app.

How:

• Added .maestro/flows/sos.yaml with three scenarios:
  1. Full activation — hold SOS button → countdown → contacts notified
  2. Cancel during countdown — hold → confirm dialog visible → tap cancel → no SOS sent
  3. No contacts configured — SOS fires → shows "Add Emergency Contact" prompt
• Mock scripts in .maestro/scripts/ intercept POST /api/emergency/sos and GET /api/emergency/contacts so no real alerts are dispatched during CI
• Run with: maestro test .maestro/flows/sos.yaml