fix(backend): detect ledger gaps via a contiguous-cursor checkpoint table

[tosin-zoffun]

Summary

eventIndexer.ts advanced its cursor (indexer_state.last_ledger) whenever it was monotonic, with no record of which ledger ranges had actually been confirmed scanned by the RPC. A restart that resumes from a stale or clamped starting point could silently skip a range of ledgers with no error — empty ledgers legitimately produce no events, so the old code had no way to distinguish "empty" from "skipped."

This adds a checkpoint table so the indexer can detect that kind of gap and flag it, and gates the one consumer (defaultChecker.ts) that draws conclusions from indexed events on there being no unresolved gap for the loan manager contract.

• New ledger_checkpoints table (contract, range_start, range_end, status). After every poll iteration, EventIndexer.recordCheckpoint records the range that was just requested from the RPC:
  • 'verified' when range_start immediately follows the previously-recorded range's range_end.
  • 'suspect' for the gap in between when it doesn't — logged and persisted for follow-up.
• EventIndexer.getSuspectRanges() and a standalone hasUnresolvedLedgerGaps(contract) (in the new ledgerCheckpoints.ts, kept separate from eventIndexer.ts so consumers don't pull in the indexer's full dependency graph) expose this to callers.
• defaultChecker.checkOverdueLoans() now checks hasUnresolvedLedgerGaps for LOAN_MANAGER_CONTRACT_ID before evaluating overdue loans, since fetchOverdueLoanIds/fetchOverdueStats derive loan state entirely from indexed contract_events — a missed LoanRepaid or LoanDefaulted in a gap risks a false default. The run is skipped (not failed) when a gap is unresolved, returning { skipped: true, skippedReason: 'unresolved_ledger_gap' }. The gate itself fails open on a query error so a checkpoint-table hiccup alone doesn't block default checking.
• Fixed migrations/1801..._create-pause-state-table.js: it used CommonJS exports.up = async (db) => ... in an ESM package ("type": "module"), which throws ReferenceError: exports is not defined when node-pg-migrate loads it. This migration runs immediately before the new 1802..._create-ledger-checkpoints.js in file order, so it had to be fixed to verify migrate:up/migrate:down at all. Converted to the same ESM pgm-style used by every other migration in the directory — no behavior change.

Out of scope

Issue #1376's full spec covers five phases across contracts, backend, frontend, and CI. This PR is the backend detection primitive plus the one known consumer that needed gating. Left out, each being its own multi-hour, multi-file change:

• Reorg detection via a content digest (range_digest) — requires the contract-level event_seq identifier added to loan_manager/lending_pool/multisig_governance/remittance_nft first.
• Automatic backfill of a detected gap (backfillRange).
• The frontend SSE resync frame and TanStack Query cache invalidation.
• The docker-compose.yml fixture RPC and the corresponding CI job wiring.

Also unrelated and pre-existing on main (verified against a clean checkout before making any changes, not introduced by this PR):

• npm run build / npm run typecheck in backend/ already fail on transactionController.ts, pagination.ts, pauseGuard.ts, sorobanService.ts, and others.
• The migration-check job's "roll back every migration one at a time" loop breaks on migration 1800..._keyset-pagination-seq-columns.js: its down() drops an index name that doesn't match what its own up() created.

Test plan

• npm test -- --testPathPatterns="eventIndexer|defaultChecker" — 32/32 passing (7 new: gap/contiguous/first-checkpoint scenarios, getSuspectRanges, hasUnresolvedLedgerGaps happy/empty/error paths; 2 new: defaultChecker skips when a gap exists, proceeds when it doesn't).
• Full npm test — same 30 pre-existing failing suites / 15 pre-existing failing tests as a clean checkout of main (confirmed via git stash); no new failures, 9 more passing tests than baseline.
• npx eslint on every changed/added file — clean.
• Migration verified against a local Postgres 16: migrate:up → migrate:down (removes ledger_checkpoints and pause_state with no residual objects) → migrate:up again succeeds.

Closes #1376