test: run the near-miss label guards in the default tox env

[dchaudhari7177]

Closes #1082.

The hole

The default [testenv] command passes --ignore=tests/test_documentation, and only python314 clears DEACTIVATE_NOTEBOOK_AND_DOC_TESTS. So the near-miss label guards — the tests whose whole job is to redden when a stage label is renamed and a published page keeps quoting the old one — never run under a plain local tox. CI runs the 3.14 env, so the gate holds there, which is exactly what made the hole invisible.

The change

Move tests/test_documentation/test_near_miss_labels_in_docs.py to tests/test_docs_near_miss_labels.py, next to test_docs_fences.py — the precedent named in the issue, which already sits outside tests/test_documentation for this reason and says so in its docstring.

Moved whole rather than split by dependency. All 46 cases are string parsing over docs/docs/**.md plus a _STAGE_LABELS import, and the one self-test that renders a real failure builds an EvaluationResult and calls render_resolution_failure directly, so it needs no backend and no flight server. The whole file runs in 0.21s. Moving it whole also keeps one home for the label vocabulary (ALLOWED_LABELS, NEAR_MISS_BULLET_PATTERN, DOC_BULLET_STAGES, PROSE_LABEL_PAGES, LABEL_TABLE_PAGE), which a split would have had to duplicate or export.

The diff is the rename, the docstring paragraph explaining the placement, and REPO_ROOT moving from parents[2] to parents[1] because the file sits one level shallower. Nothing else changes.

Verification

Label guards collected by the default env's own selection, pytest --ignore=tests/test_documentation -m 'not notebooks':

	main	this branch
label guard cases collected	0	46

Then the rename the guard exists to catch — _STAGE_LABELS["input_data"] from "input data" to "input source", under that same selection:

FAILED test_rendered_near_miss_bullets_carry_a_current_stage_label[docs/docs/in_depth/data-access-patterns.md]
FAILED test_a_registered_page_still_renders_the_bullet_of_its_own_stage[in_depth/data-access-patterns.md]
FAILED TestProseLabelMentions::test_the_page_still_spells_the_current_label[in_depth/data-access-patterns.md-input_data]
FAILED test_the_label_table_rows_are_exactly_the_distinct_stage_labels
4 failed, 42 passed in 0.45s

The stale page and the troubleshooting label table are both named. On main the same rename is green under that selection. Source file restored afterwards — git diff main -- mloda/ is empty on this branch.

Gates on the moved file: ruff format --check --line-length 120, ruff check, mypy --strict --ignore-missing-imports all clean. No remaining references to the old path anywhere in the repo.

One note on the issue text

The issue mentions "the two self-tests that render a real failure and drive GlobalFilter.warn_on_unmatched_filters". In this file there is one such self-test (test_the_scanner_pattern_matches_a_rendered_near_miss_bullet); nothing here touches warn_on_unmatched_filters. It made no difference to the outcome, since the render is backend-free either way and the file moved whole — flagging it in case a second guard was meant to exist and is missing.

Rebased onto current main (9caea51d), which picked up the parents[2] anchoring and the relative parametrize ids from #937 in the meantime; this branch keeps those as they are.

[TomKaltofen]

@dchaudhari7177 Please rebase your PR.