test(fixtures): factory_boy factories + delegate make_* helpers (#1272)

[jonfroehlich]

First PR slice of #1272 (testing leverage — realistic test data). Introduces factory_boy + Faker fixtures as the single source of truth for building model instances, and rewrites the DatabaseTestCase.make_* helpers as thin wrappers over them.

What's in this slice

• factory_boy==3.3.3 + Faker==40.23.0 added to requirements.txt (new Testing section).
• website/tests/factories.py — factories for Person, Project, Video, Publication, Talk, Poster, NewsItem, Award, ProjectRole.
• website/tests/base.py — make_* helpers now delegate to the factories, preserving their original keyword API (notably the year=YYYY shorthand and with_thumbnail) so the ~200 existing call sites stay unchanged.
• website/tests/test_factories.py — 10 tests guarding the factories themselves.
• _GIF_1PX moved to factories.py (single home); the one importer (test_easter_egg_picker.py) repointed.

Two design decisions

• Filesystem-light by default. Factory file fields use tiny in-memory uploads (a 1×1 GIF, a stub PDF) — same approach base.py already used. Keeps the DB suite fast and skips Person.save()'s Star Wars fallback (which reads a real file off disk). Curated real seed_media/ is deferred to the layer-2 follow-up where exercising real thumbnail/crop/PDF paths actually matters.
• No auto-authors. Publication/Talk/Poster factories create authorless artifacts unless authors=[...] is passed. This keeps the existing make_publication call sites behavior-identical (they add authors explicitly) and preserves authorless-path coverage. The issue's example showed auto-3-authors; diverging is documented in the factory docstring. seed_dev_data can pass authors=PersonFactory.create_batch(n).

Why

Foundation for #1278 item 5 (backfill high-risk untested code) and #1272 layer 2 (realistic local dev data). Building the connected ProjectRole/M2M/SortedManyToManyField graph is now one line instead of twenty, so future test-writing and data-seeding get much cheaper. No production behavior changes — pure test infrastructure.

Testing

• Full suite green: 211 passing (was 201) via python manage.py test website --settings=makeabilitylab.settings_test.
• Factory tests run 3× to confirm no Faker-driven flakiness (Faker is unseeded; no assertion keys off a generated value).
• CI installs factory_boy/Faker fresh from requirements.txt.

Follow-ups (separate PRs, per the issue)

• seed_dev_data management command + curated seed_media/
• scripts/pull_prod_subset.sh
• Testing automation roadmap: CI, coverage, and the path beyond localhost tests #1278 items 4 (coverage), 5 (backfill), 6 (Pa11y in CI)

Note for local dev

A requirements.txt change needs a rebuild to persist locally: ./run-docker-local-dev.sh --build. (I pip install'd into the running container to validate this branch.)

🤖 Generated with Claude Code

[jonfroehlich]

Merge follow-up: this branch was cut before the docs were touched, so on merge one line in CONTRIBUTING.md (~L267, "Running the Test Suite") goes stale — it says the make_* helpers are "built on plain Model.objects.create()", which is exactly what this PR changes (they now delegate to the factories). Update that sentence to mention website/tests/factories.py when merging (or in a quick follow-up), and consider adding a one-liner pointing contributors at the factories for richer fixtures.