perf: Add an ID-only response for rows

[priethor]

What

Part of #298.

The rows endpoint now accepts shape=ids, which returns the same ordered and filtered result as a full request, but with IDs and pagination totals instead of row data. It is available in edit context, supports up to 1,000 IDs per page, and leaves full responses as the default. Requested calculations still cover the entire result, not just the current page.

Core REST row responses now return the same cover and field data as the rows endpoint.

Why

This separates the row list from the row records before moving those records to core-data. The UI does not use shape=ids yet, but projected requests such as Gutenberg link suggestions already do less work.

How

ID-only requests use the field-value index when possible and fall back to postmeta otherwise. Core REST requests honor _fields, so they skip any row enrichment the caller did not request. The backend benchmark covers both row shapes and the link-suggestion projection.

Testing Instructions

1. Open a collection with typed fields, relations, rollups, and covers. Search, sort, or filter it, then open and reload a row. Confirm its values and cover still appear correctly.
2. Search for a Cortext document or row in Gutenberg's link picker. Confirm that the title and link are correct.
3. Using an authenticated REST client, request the same collection with shape=full and shape=ids. Confirm that both return the same order and totals, and that the ID response contains only ids, total, totalPages, and any requested calculations.
4. Add a calculations[field-X] operation and use a small per_page value. Confirm that both shapes return the same calculation for the entire filtered result, not just the current page.
5. Request 1,000 IDs in edit context, then repeat the request with context=view. Confirm that the first succeeds and the second is rejected.