fix: green the backend pipeline and re-arm CI enforcement

[godamongstmen897]

Repairs the fallout from merging the 25-PR backlog, and re-arms the CI gate that #397 temporarily lifted.

	before	after
tsc --noEmit	98 errors	0
npm test	31 failing / 87 suites	1549 passing, 87/87 suites
npm run build	—	passes

The part worth reviewing carefully

Most of this is not "fix the type error" — it is restoring feature code that successive merges silently deleted. Nothing caught it, because each merge reverted the implementation and its tests together, so the suite stayed green about code that no longer existed.

Restored:

• indexer_metrics_collector — schema verification hooks (Add migration verification hooks to indexer_metrics_collector #340), RPC health check with backoff retry (Build exponential backoff retry strategies on indexer_metrics_collector #334), resetIndexerMetricsCollectorState
• sqlite_vacuum_cleaner — index accessors (Configure SQLite index structures for sqlite_vacuum_cleaner #344), opening diagnostics boundary of a cleanup cycle
• ledger_range_tracker — schema verification incl. migration-gap detection
• indexer_runner — IndexerRunnerFailureMonitor and its throttle wiring (Optimize poller throttling parameters in indexer_runner #256)
• poller — historical range import (Configure dynamic historical sync ranges in indexer_runner #254), poll diagnostics (Write polling diagnostics logs for rpc_poller_client #270), dynamic poll interval
• routes/jobs — partial-release cache key and its cache-hit path (Implement Node-Cache lookup values for POST /api/jobs/:contractId/milestones/:index/partial-release #116); the cache was being written and deleted but never read

Merge damage repaired

• db.ts had SCHEMA_MANAGER_INDEXES declared four times and insertHistoricalEventBatch three times. Kept the array form and the range-validating overload every caller actually uses.
• poller.ts referenced two architectures at once — a bare Server and rpcClient. Settled on the retry client, exposing rpcServer so fetchEventsWithRetry applies its own policy rather than nesting retries 5×3.
• runMigrations is now atomic across pending migrations and reports failures to the schema-manager monitor. Bootstrap failures previously reached neither — they threw straight past the monitor, leaving a half-migrated database.
• database_writer_pool — the queue's persisted-key cache now clears on reset. That cache describes one database, so it survived a setDb swap and made the pool skip inserts for rows the new database had never seen.

New migration

Migration 8 adds the event_type lookup indexes the topic filter needs. Migration 3 indexed (contract_id, event_type), which cannot serve a bare event_type predicate — event_type is not the leading column.

Test changes, and why they are not cheating

Three expectations were stale rather than wrong. Each now asserts the property the ticket cared about instead of an incidental detail:

• two index assertions named a specific index, before later migrations gave the planner a better-suited one — now assert the lookup is index-backed rather than a full scan
• a migration list hardcoded [5, 6, 7] — now derived from getShippedMigrationVersions()

One genuine conflict needed a design call: two suites demanded different things from the same poll. One asserts exactly one debug line per poll; the other greps for a stall-diagnostics line. Two separate lines cannot satisfy both, so a poll now emits one structured diagnostics record carrying both, with the first line bearing a stall window exempt from the throttle.

CI

continue-on-error removed from Type check, Test and Build — the workflow is byte-for-byte its original enforcing form again.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.