fix: remove deleted dag/captive-core imports and wire persistent cursor into indexer worker

[Abd-Standard]

Closes #283
Closes #280

---

Issue #283 — Remove deleted dag and captive-core module references

Problem

lib/stellar/indexer.ts imported reconstructDagFromMetaXdr from
../dag/engine and the ExecutionDag type from ../dag/types. Both files
were deleted in a prior cleanup commit, causing a build failure. The same
commit also deleted lib/stellar/captive-core.ts, leaving
CaptiveCoreSupervisorOptions, CaptiveCoreControls, and
startCaptiveCoreIndexer referenced in startResilientEventIngestion with
no backing module.

Changes

• Remove the two dead dag import statements
• Remove the onDag field from StreamingIndexerOptions
• Remove the DAG reconstruction block from the Horizon streaming callback
• Remove the captiveCore option from ResilientStreamingOptions and the
  startCaptiveCoreIndexer call from startResilientEventIngestion
• Add the missing IngestionStateStore import from ./ingestion-state and
  re-export createMemoryIngestionStateStore / createFileIngestionStateStore
  so existing consumers (server.ts, tests) require no import-path changes
• Add stateStore and coldStartLookbackLedgers to StreamingIndexerOptions
  to satisfy the call-sites that were already using them
• Fix duplicate paginationCursor property in the poll loop
• Remove the now-invalid captiveCore option from server.ts
• Update lib/stellar/__tests__/indexer.test.ts to drop the captive-core
  test path (which depended on the deleted module)

Result

File compiles cleanly. All 7 existing indexer unit tests pass.

---

Issue #280 — Wire persistent cursor into the standalone indexer worker

Problem

src/worker/indexer.ts never called getCursor() on startup or
updateCursor() after processing batches, so every restart re-indexed from
ledger 0 — wasting RPC quota and delaying the dashboard while backfilling.
lib/stellar/indexer-persistent.ts and lib/db/utils.ts already provided
the full cursor infrastructure; it just wasn't wired up.

Changes

• Startup read: call getCursor() on boot; resume from the stored ledger.
  Fall back to the START_LEDGER environment variable (default 0) on first
  run when no cursor row exists yet.
• Mid-run write (streaming mode): call updateCursor(rawEvent.ledger)
  after every successfully handled event so the cursor advances continuously.
• Mid-run write (polling mode): call updateCursor(cursor.lastLedger)
  after every successfully processed batch, not only at shutdown.
• Graceful SIGTERM / SIGINT: stop() now (1) signals the underlying
  indexer to stop producing, (2) awaits the in-flight batch via a Promise
  sentinel so no partially-processed ledger is ever lost, (3) writes the
  final cursor to the database, then (4) tears down Redis and exits.
• Convert SorobanRpc.Api.EventResponse → RawEvent via
  eventResponseToRawEvent in polling mode before publishing.
• Document START_LEDGER in .env.example with usage notes.

Tests added (src/worker/__tests__/indexer.test.ts)

11 unit tests with a fully mocked Prisma client covering all three scenarios
required by the issue:

Suite	Tests
Startup cursor read	returns 0 on first run; returns stored ledger on resume; queries correct table/id
Mid-run cursor write	upserts after each batch independently; creates row on first write; updates on subsequent writes
Graceful shutdown	persists cursor before stop() resolves; skips write when ledger is 0; in-flight batch awaited before final write; restarts resume from persisted ledger

Result

18/18 tests pass (11 new + 7 pre-existing).

---

What was tested

• lib/stellar/__tests__/indexer.test.ts — 7/7 ✅
• src/worker/__tests__/indexer.test.ts — 11/11 ✅
• TypeScript compilation: tsc --noEmit --project tsconfig.server.json — no
  errors in any file touched by this PR

Note: 11 pre-existing test failures exist in lib/translator,
lib/resilience, and lib/stellar/__tests__/ingestion-pool.test.ts.
These were already failing on main before this branch and are unrelated
to the changes here.