perf: index list fields for store/API/MCP listings (entry_list_values) #110

Why: RFC-0042's 10,000-entry benchmark (#105, pnpm --filter @cloudhal/astro-renderer bench:lists) puts the store, API and MCP listing path above the threshold the RFC set (open question 10: more than 50 ms CPU or more than 10,000 rows read per list):

The public site is fine. Studio lists, GET /entries?where=… and MCP list_entries read the whole collection on every call.

Plan (RFC-0042 §4, "side table"):

• An entry_list_values(site_id, entry_id, field, kind, num, text) table, written in the entry write batch (rule 4).
• Only fields in a collection's listFields are indexed.
• Expression indexes on (site_id, field, num) / (site_id, field, text).
• queryContentEntries pushes where/orderBy/paging down to SQL and keeps the shared evaluator as the oracle.
• Backfill as a resumable migration.
• Field operations (rename, retype, remove) keep the table in step.

Done when: the benchmark's store path is under the threshold on sqlite and D1, results are identical to the in-memory evaluator (property test), and the existing suites pass.