Paginate delete blockers to prevent OOM

[foozleface]

Fixes #7515
Contributed by @foozleface

The delete_blockers endpoint used Django's Collector.collect() which loads ALL related objects into memory before returning. For a Taxon like "Life" with 900K+ determinations, this causes out-of-memory kills. This PR replaces the Collector with targeted COUNT queries and limited ID lists per PROTECT relationship, keeping memory bounded regardless of how many blocking rows exist.

Implementation

• Replace Collector.collect() in _collect_delete_blockers with iteration over obj._meta.related_objects, filtering for protect_with_blockers on_delete
• For each PROTECT relationship, run a COUNT query and fetch at most id_limit (default 100) primary keys
• Add total_count field to the response so the frontend knows the full count even when IDs are capped
• Update frontend test fixtures to match the new isstrict: undefined serialization
• Add comprehensive backend tests for the pagination behavior

Testing instructions

• Navigate to a record with many dependent records (e.g., a high-level Taxon node) and attempt to delete it
• Verify the delete blocker dialog shows the correct count and a reasonable number of linked record IDs
• Confirm no OOM or timeout on records with very large numbers of blockers
• Run the test suite: python manage.py test specifyweb.backend.delete_blockers.tests.test_pagination