Eliminate N+1 queries on Project & Person admin changelists (#1346, Phase 2)

[jonfroehlich]

Phase 2 of #1346 (admin changelist audit). The two heaviest admin list pages were doing per-row DB work that scaled with the number of rows:

• Projects — ~10 DB-touching callable columns (get_banner_count, the member counts, get_most_recent_artifact_* ×2, the contributor union…) × the default 100 rows/page = 1,000+ queries for one list view.
• People — ~14 columns, nearly all funneling through per-row position lookups (.latest()/.earliest()), plus the PositionRoleListFilter looping every Person on every load.

Both are now roughly constant-query regardless of row count, and the count columns became sortable as a bonus.

Project (project_admin.py)

• get_queryset() annotates every count as an independent scalar subquery (related_count_subquery), so the joins don't multiply each other (stacking Count() over several M2M relations both blows up rows and risks wrong counts). Covers pubs, talks, videos, banners, distinct people, current/past members, and the role∪publication-author Contributors union (kept per your call, computed as one correlated subquery).
• Most-recent-artifact date and type derived from three max-date subqueries combined with Greatest().
• Columns repointed at the annotations; the model methods stay for the public site/detail views. list_per_page = 50.

Person (person.py + person_admin.py + position_role_list_filter.py)

• get_latest_position / get_earliest_position now pick from position_set.all() in Python instead of .latest()/.earliest(), so a single prefetch resolves them — and with them every "current title/role/date/duration", is_active, and the Role filter that all funnel through them. Behavior preserved. Same prefetch-friendly rewrite for get_total_time_in_role and is_alumni_member (they used .filter() on positions).
• PersonAdmin.get_queryset() prefetches position_set + projectrole_set__project and annotates the 3 count columns; the Role filter prefetches position_set in its per-person loop. list_per_page = 50.

Shared helper related_count_subquery() added to admin/utils.py.

Tests (test_admin_perf.py)

• Correctness: each annotation / rewritten model method asserted equal to the original per-row result (e.g. pub_count == get_publication_count(), contributor_count == get_contributor_count(), most-recent date/type, the position rewrites).
• N+1 regression: renders the real admin ChangeList (filters + every column cell) and asserts the steady-state query count stays flat as rows grow 4 → 12. (Verified the fix directly: a fully-warmed 12-row People changelist is 10 queries, flat — down from ~1 + 8/row.)

Full suite green locally (362 tests, in-container with --settings=makeabilitylab.settings_test).

Merge note

Touches project_admin.py, which Phase 1 (#1347) also edits — verified they auto-merge with no conflicts in either order (different regions of the file).

Screenshots

Admin changelist sort arrows on the new count columns — to be added.

🤖 Generated with Claude Code