fix: don't remap cross-repo GitHub blob URLs to raw

[zeitlinger]

Summary

• Remove global remap of all GitHub /blob/ URLs to raw.githubusercontent.com
• Keep same-repo remaps (build_remap_args) and issue/PR comment anchor stripping

Problem

The global remap breaks when a /blob/ URL points to a directory (e.g. github.com/org/repo/blob/main/some-dir). GitHub handles these via redirect to /tree/, but raw.githubusercontent.com returns 404 for directories.

Found in open-telemetry/opentelemetry-java-contrib#2654 where https://github.com/open-telemetry/opentelemetry-java-examples/blob/main/javaagent (a directory) was being remapped and failing.

Test plan

• Verify link checking passes on repos with cross-repo /blob/ directory links
• Verify same-repo remaps still work (rate limiting avoidance)

[Copilot]

Pull request overview

This PR adjusts the link-checking remap logic used by lint:links to avoid incorrectly rewriting cross-repo GitHub /blob/ URLs to raw.githubusercontent.com, which can cause false 404s when the /blob/ URL points to a directory (GitHub redirects those to /tree/, but raw does not).

Changes:

• Removes the global remap of all GitHub /blob/ URLs to raw.githubusercontent.com.
• Keeps same-repo remaps via build_remap_args() and retains issue/PR comment anchor stripping.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

build_global_github_args() no longer strips GitHub /blob/ #L… and #:~:text=… fragments for cross-repo URLs. Since lychee runs with include_fragments=true (see .github/config/lychee.toml) and the repo contains external #:~:text links (e.g. tests/test-links.md), this will reintroduce deterministic false failures because Scroll-to-Text fragments are not present in static HTML. Consider keeping fragment-stripping remaps for /blob/ URLs (rewrite to the same https://github.com/<owner>/<repo>/blob/<ref>/<path> without the fragment) while avoiding any remap to raw.githubusercontent.com, which preserves the directory /blob/ → /tree/ redirect behavior that motivated this change.

Suggested change

	echo '^https://github.com/([^/]+/[^/]+)/(issues|pull)/([0-9]+)#issuecomment-.*$ https://github.com/$1/$2/$3'
	echo '^https://github.com/([^/]+/[^/]+)/(issues|pull)/([0-9]+)#issuecomment-.*$ https://github.com/$1/$2/$3'
	# Cross-repo /blob/ URLs with JS-rendered fragments:
	# - Line anchors (#L123, #L10-L20, etc.)
	# - Scroll-to-Text fragments (#:~:text=...)
	# Lychee can't verify these when include_fragments=true, and the
	# fragments don't exist in static HTML. Strip the fragment while
	# keeping the URL on github.com (no remap to raw.githubusercontent.com).
	echo "--remap"
	# Strip line-number anchors on blob URLs
	# shellcheck disable=SC2016 # single quotes are intentional: these are regex capture groups, not shell vars
	echo '^https://github.com/([^/]+/[^/]+)/blob/([^#]+)#L[0-9]+.*$ https://github.com/$1/blob/$2'
	echo "--remap"
	# Strip Scroll-to-Text fragments on blob URLs
	# shellcheck disable=SC2016 # single quotes are intentional: these are regex capture groups, not shell vars
	echo '^https://github.com/([^/]+/[^/]+)/blob/([^#]+)#:~:text=.*$ https://github.com/$1/blob/$2'

[zeitlinger]

After investigating all 170 cross-repo /blob/ URLs across the 11 consuming repos, only 4 point to directories. GitHub redirects /blob/ → /tree/ for directories (301), so they work — but raw.githubusercontent.com returns 400/404.

Rather than adding the extension heuristic here, the simpler fix is to correct those 4 URLs upstream to use /tree/. The raw remap being stricter is a feature — it surfaces genuinely wrong URL types.

Fixed the 3 URLs causing CI failures in java-contrib#2654: c309e075. Keeping this PR open until we confirm that resolves the issue, then closing.

[zeitlinger]

Closing — the fix is to correct the 4 upstream directory URLs to use /tree/ instead of /blob/, not to add a heuristic in flint. See #103 (comment) for details.