fix(indexer): wrap event handlers in Prisma transaction to prevent poison-pill events (#168)

[trenysx]

Summary

Closes #168

Problem: Event handlers in persistEvent perform multiple separate prisma.* calls without a transaction wrapper. If a handler fails mid-way (e.g., investment.create succeeds but campaign.update throws), the partial writes remain committed while the idempotency-marking Transaction row is never written. On the next poll, the event is reprocessed but immediately fails on the unique constraint of the already-existing partial row — a permanent "poison pill" that spams error logs forever.

Solution: Wrap each handler's writes + the final Transaction row insert in a single prisma.(async (tx) => { ... }) so a partial failure rolls back cleanly and the event can be retried from a consistent state.

Changes in server/src/indexer/parsers/event-parser.service.ts:

• persistEvent now uses $transaction to atomically execute the handler + Transaction row insert.
• All 16 handle* methods + ensureUser now accept a transaction client ( x) parameter and use x.* instead of his.prisma.*.
• Administrative/audit-only events (registry admin events) still fall through to the Transaction row creation inside the same transaction.

Tests added in event-parser.service.spec.ts:

1. ransaction atomicity (poison-pill prevention) suite with two tests:

• Simulates a mid-handler failure: verifies Transaction row is NOT created (event not marked processed), then retries the same event successfully with full commit.
• Verifies event is only marked processed after the full transaction commits.

All 47 unit tests pass; sc --noEmit and eslint are clean.