test(ingestion): reorg simulation deduplication test

[Depo-dev]

Summary

Adds a reorg simulation integration test that validates the ingestion layer is fully idempotent under replay conditions (Issue #79).

• tests/integration/reorg.test.ts: 12 deterministic tests across 4 suites:
  1. Normal ingest: all records inserted on first pass
  2. Full replay: 0 new records on second identical pass, store size unchanged, no duplicates
  3. Skipped-ledger gap simulation: detects missing ledger, backfills without duplicating existing records, all data present after resume, consistent after aggressive overlapping replays
  4. upsertTransfers contract: always passes skipDuplicates:true, skips Prisma call for empty input, returns correct inserted count
• @prisma/client is mocked at the PrismaClient constructor level so the real upsertTransfers code path is exercised with an in-memory store that mirrors the Postgres @unique(eventId) constraint
• tsconfig.test.json: extends root tsconfig with rootDir:. so tests/ compiles alongside src/ with full jest types
• package.json: adds tests/ to jest roots + testMatch; uses tsconfig.test.json via transform config

Also fixes two pre-existing build errors now surfaced by the stricter test compilation:

• src/db.ts: add local toDisplayAmount helper (was used but never imported)
• src/api.ts: fix type assertions on OData-projected transfer fields in CSV export

Acceptance criteria

• Deterministic deduplication works (replay inserts 0 records)
• Replay-safe ingestion confirmed (all records present, none missing after gap + resume)
• No regression in existing ingestion pipeline (58 tests pass across 5 suites)

Test plan

• Run: npx jest tests/integration/reorg.test.ts -- confirm 12/12 pass
• Run: npx jest -- confirm no regression in existing passing test suites

Closes #79

[drips-wave]

@Depo-dev Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits

[Miracle656]

Tight, well-targeted dedup proof.

• Mock strategy: replacing PrismaClient at constructor level lets the real upsertTransfers code path run against an in-memory Map<string, TransferRecord> that mirrors the Postgres @unique(eventId) constraint. Same code path the production indexer uses — exactly the right contract test.
• _createManyCalls spy confirms upsertTransfers always invokes createMany with skipDuplicates: true (a structural guarantee, not just an observed behavior).
• Skipped-ledger gap test is the sharpest one: ingest ledgers 1–3 and 5 (4 missing), then replay 1–5 → only the previously-missing ledger 4 events get added, no duplicates on the rest. That's the actual #79 acceptance criterion ("no dup rows after replay" + "no data loss after skip-then-resume").
• The Prisma namespace stub (sql, join, empty) prevents the mock from breaking on raw-SQL helpers used elsewhere in db.ts.

The two side-fixes in src/db.ts and src/api.ts are real compile errors that surfaced once the stricter test tsconfig pulled them into scope:

• toDisplayAmount was used in db.ts without an import — adding a local helper is the minimal fix.
• The OData CSV-export type assertion was unsound after #92 landed; tightening it is the right call.

Both fixes are scoped to making the test pipeline green, transparent in the PR body. Acceptable scope creep.

Heads-up to next contributor: this PR adds tsconfig.test.json and the same roots/testMatch/transform jest config that #88 (still awaiting rebase) also wants to add. arandomogg will hit a conflict on those two files when they rebase — straightforward "keep both" resolution.

Merging. Closes #79.