Repair pre-commit, drop pixi.lock, point .gitattributes at uv.lock

[xhan97]

Three related bits of housekeeping. The pre-commit part is the substantive one.

.pre-commit-config.yaml was a trap

It ran ruff-format while CI enforces black, and the two disagree:

$ black --check .          # what CI runs
69 files would be left unchanged.

$ ruff format --check .    # what the hook did
4 files would be reformatted

$ ruff format . && black --check .
5 files reformatted
Oh no! 💥 💔 💥  1 file would be reformatted

Any contributor who ran pre-commit install would have had their commits reformatted into a state that fails the lint job in CI. It went unnoticed only because nothing runs it — no git hook installed, no workflow invoking it — even though pre-commit>=4.3.0 sits in the lint dependency group.

Fixed by making formatting black's alone, running after ruff-check so the formatter has the last word.

The revs had also rotted:

hook	was	now
ruff-pre-commit	v0.12.1	v0.16.3 (matches the project)
prettier	pre-commit/mirrors-prettier	rbubley/mirrors-prettier v3.9.6
validate-pyproject	v0.24.1	v0.25
codespell	v2.4.1	v2.4.3
black	absent	26.5.1 (matches the pin)

pre-commit/mirrors-prettier is archived upstream. A pre-commit ecosystem is added to dependabot.yml so these stop drifting silently — that drift is how it got here.

Deleting the config would have been simpler, but it would have dropped codespell, validate-pyproject, and prettier, none of which CI runs. [tool.codespell] in pyproject.toml is read by nothing else.

What codespell found once it ran

Eleven hits. Five were false positives:

• lamda ×2 — the LAMDA lab's domain in author emails (mayx@lamda.nju.edu.cn)
• anc ×3 — a local variable for "ancestors" in _inode.py

Those are now in ignore-words-list with a comment explaining why. The other six are real and corrected, two of them user-facing:

• Subsequnce → Subsequence in IKTOD's class docstring
• lables → labels in a doctest example in _ikgod.py (last line of the example, nothing references it)
• dimention, Exsiting, commiting, continous

Churn worth calling out

prettier normalized indentation in three workflow files and removed a blank line from mkdocs.yml — the first time it has run over them. This is unrelated to the stated purpose of the PR, so:

• Parsed YAML verified identical before and after for all three workflows
• mkdocs.yml uses custom tags so it cannot be safe-loaded; its only change is one deleted blank line, and mkdocs build --strict still passes

pixi.lock and .gitattributes

pixi.lock is 438K of lockfile with no pixi.toml and no [tool.pixi] section anywhere — orphaned since the move to uv in November 2025.

.gitattributes existed only to mark it generated, duplicated across two identical lines:

pixi.lock linguist-language=YAML linguist-generated=true
pixi.lock linguist-language=YAML linguist-generated=true

Rather than delete it, it now does the same for uv.lock, which collapses that file in GitHub diffs and keeps it out of language stats. Every dependency PR in this repo buries its real change under hundreds of lock lines; this fixes that.

Verification

All six hooks pass on --all-files. black --check, ruff check, and uv lock --check are clean afterwards — the round-trip that was previously broken.