Improve the ListUsageController to not process all the rows at the same time

Problem/Motivation

The usage list page loaded every source entity referencing the viewed entity, even entities not shown on the current page. For entities with many references, this loaded and processed data that pagination then discarded.

Proposed resolution

Old pipeline

ListUsageController::getRows() called listSources(), loaded every source entity with loadMultiple(), and built a row for each one. Pagination sliced the finished list afterwards.

New pipeline

• getDisplayableEntries() calls listSources() without loading entities.
• filterDisplayableEntries() checks entity existence and, for inline blocks, the "reusable" field, with entity queries instead of full loads. This gives the exact ordered row list and pager total.
• listUsagePage() slices this list to the requested page. buildRows() loads only that page's source entities with loadMultiple(), grouped by type, and builds the rows.

Trade-offs

The existence and field checks have moved to filterDisplayableEntries() instead buildRows() (orphaned usage records, soft-deleted inline blocks). This opens a tiny gap where things can change while building the usage records. Given that this will only result in displaying a row that would have been just a moment before this feels acceptable.

Remaining tasks

User interface changes

API changes

Data model changes