[H5] "Torch" literal in lucid/models/vision/densenet/{_pretrained,_model}.py

[ChanLumerico]

Rule: H5 — No "torch" / "PyTorch" in source

H5 (CONTRIBUTING.md §3) bans the words torch / PyTorch from source code, comments, docstrings, and error messages anywhere under lucid/ — the only sanctioned exception is lucid/test/_fixtures/ref_framework.py. The neutral phrasing to use is "reference framework" (the surrounding text already uses "reference vision library").

These four occurrences were introduced recently in commit cf7dcc4 ("fix(models): close every confirmed finding in the model-zoo audit"), so they are a fresh regression, not part of the historical backlog. The detection command grep -ri "torch\|pytorch" lucid/ --include="*.py" --exclude-dir=test flags them.

Sites (path:line → offending text)

#	Location	Text
1	lucid/models/vision/densenet/_pretrained.py:24	comment: "The provenance is the authors' own released Torch models"
2	lucid/models/vision/densenet/_pretrained.py:507	docstring: "It comes from the authors' own released Torch models"
3	lucid/models/vision/densenet/_pretrained.py:572	docstring: "it comes from the authors' own released Torch models."
4	lucid/models/vision/densenet/_model.py:68	comment: "as the authors' original Torch code does."

Why it matters

H5 is enforced in CI; a PR carrying these would be rejected. Beyond the rule, "Torch" here names the reference framework that Lucid is deliberately kept free of in its own sources.

Suggested fix

Replace "Torch" with a neutral term such as "the reference framework" / "the reference implementation" in all four spots, e.g. "the authors' own released reference-framework models". No behavioral change; comments/docstrings only.