[rudycelekli]

Long stylesheets are repeatedly scanned in full while checking canvas-level zoom: the rule loop calls stripCssComments(style.content) for each matched CSS rule and once more when the loop ends. Strip once per stylesheet and reuse the same string for the unchanged regex loop.

Fixes #5681.

Validation:

• Ran the actual stylesheet scanning function with the actual comment stripper under Node 24.19.0. For a 1,000-rule stylesheet, full comment-stripping calls fall from 1,001 to 1, with identical matched rule headers/bodies. Multi-stylesheet, comment/string, empty stylesheet and no-stylesheet controls preserve identical inputs to the downstream classifier. No elapsed-time speedup claim is made.
• Repository-config Oxlint 1.64.0 (lockfile version), Oxfmt 0.41.0, comment citations and comment ratchet pass. Citation checking reports one non-blocking pre-existing warning about diagnoseDerivedEnd.

Validation limitations: this characterization captures inputs to the downstream classifier rather than running the complete linter. The prescribed workspace build was attempted and failed because tsup is unavailable. Full linter suite, project-level CLI reproduction and workspace typecheck have not passed locally.

AI-assisted implementation and validation.