Keep OGM harvests synchronized with search

[ewlarson]

Problem and fix

OGM harvests updated PostgreSQL but never refreshed Elasticsearch. Production held 321,304 published resources while unfiltered search returned only 136,942 indexed records.

Synchronize each harvested repository into the live index before reporting success. The helper uses keyset pagination, bulk writes, and prefetched enrichments in an isolated process; it refreshes search and suggestion caches afterward. Unpublished database records are removed from search, and indexing failures make the harvest visibly fail. The live index and PostgreSQL data are preserved.

Prepare v0.10.1 and document manual reconciliation and the distinction between repository memberships and unique records. Search also includes published records without repository tags.

Validation

• 12 focused tests passed, including pagination, unpublished-record removal, cache invalidation, partial-index failure, harvest failure propagation, and dashboard/API rendering.
• Ruff and whitespace checks passed.
• A separate production repair index indexed all 321,304 published records with zero errors and an exact database count match. Live search is being reconciled from that validated index.