feat: decouple ingestion pipeline jobs

[mdiapenabc]

Description

Splits the monolithic data ingestion pipeline into three independently runnable jobs. Previously, INGEST_DATA chained into eligibility, which chained into auto-batching and ICM sync-back. Each step now runs as a standalone job with its own entrypoint, so they can be triggered independently from OpenShift or manually.

What Changes

• INGEST_DATA now stops at staging tables: fetches ICM + MIS data in parallel, no eligibility chain
• RUN_ELIGIBILITY runs independently: upserts master contacts table, then syncs flagged contacts back to ICM
• AUTO_BATCH is a new independent job: inds all eligible / not_eligible_in_pay contacts and adds them to the pending batch, then syncs back to ICM
• New entrypoint: backend/src/jobs/entrypoints/auto-batch.ts (runnable as job:auto-batch)
• Fixed eligibility threshold to use last RUN_ELIGIBILITY success timestamp instead of INGEST_DATA
• Removed dead autoBatchContacts() method and autoBatched field from EligibilityService (logic now lives exclusively in AutoBatchService)