Harden Elasticsearch indexing and search for larger module corpora

Problem

The timeout increase in #4463 is an immediate mitigation, but the scraper/search integration still has several size-related assumptions that are now brittle with the current corpus size.

Observed production context:

• modules_v2 has about 21,096 documents.
• moduleInfo bulk indexing is sent as one large request, about 18.5 MB.
• ElasticPersist.getModuleCodes() fetches at most 20,000 docs in a single search.
• modules_v2 has index.max_result_window = 20000.
• Frontend pagination/random-course paths can request beyond the ES result window and receive HTTP 400.
• createIndex() assumes a specific error shape and can throw Cannot read properties of undefined (reading 'type') for other ES errors.

Proposed work

• Chunk scraper moduleInfo bulk writes by document count and/or byte size.
• Add retry/backoff for transient ES timeout/connection failures.
• Replace getModuleCodes() single size: 20000 search with scroll/search-after so cleanup sees all existing documents.
• Fix frontend deep pagination/random-course behavior so it does not request beyond index.max_result_window, or move those paths to a cursor/search-after-safe design.
• Harden ES error handling in createIndex() and delete paths so unexpected ES/network errors are logged clearly without masking the original cause.

Notes

This is intentionally separate from the timeout mitigation PR so the production scraper can be unblocked first.

Start by tracing scraper moduleInfo bulk indexing, ElasticPersist.getModuleCodes(), createIndex(), delete paths, and the frontend pagination/random-course paths named in the issue. Verify the current Elasticsearch request limits and error behavior before splitting the work into focused changes. Done means large corpora index and clean up reliably, deep pagination avoids result-window failures, and unexpected Elasticsearch errors retain clear original causes.