feat: persist bet-store with Prisma so bets survive restarts (#519)

[Richardkingz2019]

Summary

Closes #519 — Persist bet-store beyond process memory.

The bet store lived entirely in process-local Maps (src/data/bet-store.ts), so the bet audit trail was lost on every deploy/restart and was invisible to other instances sharing the database. This PR backs the store with a durable Prisma BetRecord table so demo bets survive restarts and multi-instance demos, while keeping the pure in-memory backend for true mock mode.

What changed

Persistence model

• New BetRecord Prisma model (prisma/schema.prisma) + migration 20260827000000_add_bet_record:
  • Money columns (amount, predictedPrice) use Decimal(20, 8) / Decimal(18, 8) per the repo's monetary-precision rule.
  • Indexed on address, roundId, status, timestamp for the admin/audit query paths.
• src/lib/prisma.ts gains an in-memory betRecord stub so unit tests exercise the same Prisma-shaped API as production without a live DB.

Bet-store refactor (src/data/bet-store.ts)

• New BetStoreBackend interface with two implementations:
  • InMemoryBetStore — process-local Maps; used when DATA_STORE=memory (auto-derived from DATA_MODE=mock). Nothing survives a restart.
  • PrismaBetStore — bets persisted in BetRecord; used for DATA_STORE=postgres (the DATA_MODE=live default). Bets survive restarts and are shared across instances.
• The BetStore facade resolves the backend lazily per call (resolveBetStoreKind()), so:
  • Mock mode keeps working with zero DB access.
  • Tests can flip DATA_STORE between cases without re-importing modules.
• Every backend method is async; the whole read/write API is now Promise-based.
• Id-sequence continuity: bet-{n} ids resume from the highest existing row on first write, so a restarted process continues numbering where its predecessor stopped (single-writer contract documented in code; DB-generated ids noted as the next step for true multi-writer deployments).
• getTotalBetsCount() reads from the DB so a restarted process reports the full audit trail instead of a reset counter.

Call-site migration

• BetService (src/services/bet.service.ts) — all bet-store calls awaited, including the stub → live reconcile path and failure handling.
• GET /api/bets/reconciliation and GET /api/bets/:id (src/routes/bets.routes.ts) — now async, summary computed in parallel with the bet list.
• InMemoryRoundRepository / InMemoryStatsRepository (src/repositories/in-memory.repositories.ts) — awaited.

Operational hardening (found while making CI green)

• src/config/node-version.ts — fail-fast Node ≥ 22 gate imported as the very first side effect of src/index.ts, so an unsupported runtime dies with a clear message instead of an opaque ERR_REQUIRE_ESM from @stellar/stellar-sdk. Mirrors the existing preflight check and the version-gate spec.
• .github/workflows/ci.yml — the unit job now builds first so node-version-check.spec.ts (which boots dist/index.js under Node 18) has an artifact to test.
• jest.config.ts — bet-store-persistence.spec.ts registered as an integration test.

Tests

New: src/tests/bet-store-persistence.spec.ts (integration, covers both acceptance criteria)

• Postgres mode: bets written by one store instance are readable by a fresh instance (simulated restart) — including status transitions, txHash, failure details, filters, summaries, and total count.
• Id sequence resumes across the restart boundary (bet-N shape preserved, no id reuse).
• Stub → CONFIRMED reconciliation and FAILED details survive the restart.
• Memory mode: bets stay process-local; a fresh in-memory store has no trace; the facade switches backends cleanly and never mixes stores; the full reconcile lifecycle still works.

Updated suites (await-ified / pinned to the memory backend where they assert audit-event emission, not durability):

• bet-reconciliation.spec.ts, bet-audit.spec.ts (pins DATA_STORE=memory), bet-store-decimal-precision.spec.ts, performance.spec.ts (pins DATA_STORE=memory — DB-free load suite), admin-bet-audit.spec.ts.

Pre-existing breakage fixed so the CI unit job is green again (all of these suites were red on main):

• hackathon-auth.smoke.spec.ts, http-logger-unified.spec.ts, api-contract.spec.ts — incomplete rateLimiter.middleware mocks (Route.post() requires a callback function at import).
• http-logger-unified.spec.ts — jest.isolateModules(async ...) doesn't await the callback; switched to isolateModulesAsync to stop the flaky "createHackathonApp is not a function" / teardown races.
• hackathon-logger.spec.ts — logger mock TDZ (temporal-dead-zone) crash at import; mock is now self-contained.
• hackathon-rounds.spec.ts / api-contract.spec.ts — round envelope contract updated to the shared { success, data: { source, rounds } } shape.
• admin-bet-audit.spec.ts — JWT secret set before app build (16+ char preflight requirement), role-aware user.findUnique mock.
• retention.service.spec.ts — prisma stub gains authChallenge / message / auditLog model stubs.

CI status (local verification, CI env)

Job	Status	Notes
lint (tsc --noEmit)	✅
build	✅
vendored bindings pin	✅
production-readiness scorecard	✅
contract drift (docs:verify)	✅	0 drift issues
test (unit)	✅	94/94 suites, 968 passed, coverage well above thresholds
test (integration)	⚠️ pre-existing	21 suites fail identically on main (upstream CI is red on this job for the same tests — incomplete prisma mocks, fake Soroban secrets in jest.setup.js). This PR adds bet-store-persistence.spec.ts (6/6 passing) and introduces zero new failures.
test (hackathon HTTP)	⚠️ pre-existing	Same 8 failures as main.

How to verify

npm run prisma:generate
npm run db:migrate                      # applies 20260827000000_add_bet_record
TEST_TYPE=unit npm run test:unit        # CI unit env
npm run test:integration                # DB-backed suite incl. restart-continuity tests

To see durability end-to-end: run with DATA_STORE=postgres, place a bet, restart the process, and GET /api/bets/reconciliation — the bet and its status survive.

closes #519