perf(retrieval): add content trigram index for regex grep

[suguanYang]

Summary

• add a partial pg_trgm GIN index on document_chunks.content
• build it concurrently in deployed environments
• support external-transaction migration tests
• add a schema contract test for the index definition

Why

OpenWriting grep uses PostgreSQL regex matching (content ~*). A tsvector candidate filter is not equivalent to grep regex semantics and can omit valid results. This migration accelerates the original regex predicate without changing recall.

The OpenWriting companion PR adds a regression test that protects this contract:
https://github.com/Ontos-AI/open-writing/pull/267

Verification

• uv run pytest apps/api/tests/migrations/test_schema_contract.py -q (11 passed)
• uv run ruff check (passed)
• git diff --check (passed)