Extract test set crud module and retire the split

[akwasigroch]

Purpose

Finishes the crud monolith split that #2410 started. TestSet was the last entity left in crud/__init__.py, so after this the file is empty and app/crud/ is simply one module per entity. The second half of the PR removes the now-false "split in progress" narration from the codebase.

Stacked on #2617 (Test extraction) — review and merge that one first; this PR's base is crud-extract-test, and the diff shown here is just the TestSet half.

What Changed

The move. All nine TestSet functions (get_test_set, get_test_sets, create_test_set, update_test_set, delete_test_set, get_test_set_by_nano_id_or_slug, resolve_test_set, get_test_sets_for_test, get_test_set_tests) go into crud/test_set.py, joining the bulk_delete_test_sets that already lived there. _TEST_SET_RELATED_FIELDS moves with them, so crud/explorer.py now imports it from crud.test_set rather than from the parent package. Around 90 call sites across routers, services, jobs and tests switch to test_set_crud., and the five patch("…app.crud.resolve_test_set") targets in tests/backend/services/test_test_set.py become …app.crud.test_set.resolve_test_set.

crud/__init__.py is now a single docstring line. That also takes away the names it had been re-exporting by accident, and four modules turned out to be leaning on them. services/experiment.py reached crud.get_item_detail and crud.include, and routers/experiments.py reached crud.delete_item — all three are crud_utils/query_utils names that only resolved because __init__.py happened to import them. jobs/execution/modes.py reached schemas through crud.schemas.TestConfigurationUpdate. Each now imports from where the thing actually lives. services/tool/mcp/operations.py had a crud import that nothing in the file used.

One real bug fixed. crud.delete_test_configuration does not exist anywhere in the repo — the function was dropped when test_configuration was extracted, so DELETE /test_configurations/{id} has been raising AttributeError on every call since. This adds delete_test_configuration to crud/test_configuration.py (a delete_item wrapper, matching its siblings) and points the router at it. No test covered the route, which is why it went unnoticed.

The history comes out (second commit, reviewable on its own). The "Part of the incremental split of the crud monolith: crud/__init__.py still holds…" paragraph is gone from all 29 crud module docstrings; metric.py and telemetry.py keep their "import the functions directly" guidance but lose the "Split out of crud/__init__.py" line; the monolith comment above explorer.py's _TEST_SET_RELATED_FIELDS import is rewritten. AGENTS.md's CRUD layout rule now describes the finished layout — one module per entity, __init__.py empty on purpose, a new function goes in its entity's module — instead of a migration in progress. Two stale "doesn't exist in crud.py" comments in tests and three crud.foo(...) docstring examples in utils/database_exceptions.py are reworded to the module form so the docs stop teaching the pattern that no longer works.

tests/backend/test_secret_equality.py pins its allowlist by line number, and two added imports in services/experiment.py shifted the two whitelisted content_hash comparisons from 150/206 to 152/208; the entries are repinned.

Additional Context

• Closes out the series: Extract trace CRUD out of the crud/__init__.py monolith #2410 … Extract tool and type lookup crud modules #2600 (tool, type lookup), Extract experiment crud module #2601 (experiment), Extract endpoint crud module #2602 (endpoint), Extract test crud module #2617 (test).
• No behaviour changes beyond the delete_test_configuration fix.
• Worth a look on review: the delete_test_configuration restoration is the one place this PR is not a pure move, and the route still has no test.

Testing

• Full backend suite green on this branch: 8234 passed, 62 skipped, 1 xfailed, 0 failed (5m37s). Same counts on Extract test crud module #2617's branch alone, and the 62 skips are the pre-existing ones (DEFAULT_EVALUATION_MODEL not configured, the rollback_initial_data OOM hang).
• Verified every module under rhesis.backend.app and rhesis.backend.jobs still imports, and that no crud.foo() package-level access survives anywhere in apps/ or tests/.
• ruff check and ruff format clean on all changed files.
• DELETE /test_configurations/{id} is worth exercising by hand, since nothing in the suite covers it.

[peqy]

Looks good — nice cleanup finishing the crud split, and good catch restoring delete_test_configuration.

[Improvement] Add a small router-level test for DELETE /test_configurations/{id}

Fix: add a minimal happy-path + 404 test to prevent the AttributeError regression from coming back (and to exercise tenant scoping in the route).

[peqy]

Re-reviewed latest commits — the TestSet CRUD extraction and cleanup still look solid, and the delete_test_configuration fix is wired through the router correctly.

Only remaining suggestion is the same as before: please add a minimal router-level test for DELETE /test_configurations/{id} (happy path + 404) so this regression can’t sneak back in.