fix: skip HTML comments in checkBrokenLinks

[Taegost]

What

checkBrokenLinks now strips complete <!-- ... --> HTML comment spans before scanning for links. Previously, links inside HTML comments were flagged as broken — producing false BROKEN_LINK warnings in scaffold files like patterns/INDEX.md where template examples live in comment blocks.

Unclosed <!-- (no matching -->) is left as plain text, consistent with checkIndexSync's behavior.

Line numbering fix: The initial implementation stripped comments with an empty replacement, which collapsed line numbers for any real broken links appearing after a comment block. The replacement now preserves newline characters so the line array stays the same length as the original file — reported line numbers match the actual source location.

Why

The patterns/INDEX.md template generated by mex setup includes example links inside HTML comments. After setup, running mex check on the scaffold produced 7 false BROKEN_LINK warnings from these comment examples — dropping the score to 79/100 even when the scaffold is otherwise correct.

Verified against Taegost/homelab-k8s (feat/mex-integration branch).

Changes

• src/drift/checkers/broken-link.ts — Strip <!-- ... --> from content before line processing, replacing with newline-preserving padding so line numbers remain accurate
• test/checkers.test.ts — 4 new tests: single-line comments, multi-line comments, inline comments with surrounding content, unclosed comments. Added line assertions to 2 existing tests. 2 regression tests verifying correct line numbers for broken links after single-line and multi-line HTML comments.
• CHANGELOG.md — Entry under [Unreleased]

🤖 Generated with Claude Code

The additions vs. the original:

• New "Line numbering fix" paragraph under "What" explaining the follow-up commit
• Updated the broken-link.ts bullet to mention newline-preserving padding
• Updated the checkers.test.ts bullet to mention the line assertions and regression tests

[theDakshJaitly]

Thanks for tracking down the HTML-comment false positives. The behavior change is right, but this implementation strips multi-line comments before splitting the file into lines, which changes the reported line numbers for any real broken links after a comment block.

For example, a link after:

would be reported a few lines earlier than its actual source location because the comment lines were removed before i + 1 is assigned.

Could you preserve newline positions when removing complete HTML comments, or otherwise skip comment spans without changing the line numbering? A regression test with a real broken link after a multi-line comment would cover it.

Also, the PR is currently conflicting with main, so it needs an update before merge.

[Taegost]

Thanks for tracking down the HTML-comment false positives. The behavior change is right, but this implementation strips multi-line comments before splitting the file into lines, which changes the reported line numbers for any real broken links after a comment block.

For example, a link after:

would be reported a few lines earlier than its actual source location because the comment lines were removed before i + 1 is assigned.

Could you preserve newline positions when removing complete HTML comments, or otherwise skip comment spans without changing the line numbering? A regression test with a real broken link after a multi-line comment would cover it.

Also, the PR is currently conflicting with main, so it needs an update before merge.

I updated the code to introduce \n to each line it removes, added tests, and updated the existing tests to validate the logic.