Fix filesystem aliases in vendor and Blade discovery

[shuvroroy]

Summary

• Normalize registered vendor roots once before analyzer walks compare canonical paths.
• Retain both raw and canonical vendor spellings, including after a late Composer install.
• Normalize Blade template paths against canonical view roots, with a parent fallback for deleted files.
• Add deterministic Unix regressions for analyzer vendor pruning and Blade view-name resolution through symlinks.

Why this is separate

This filesystem behavior was discovered while working on #368, but it is independent of Laravel named-resource strings. Splitting it out keeps both changes focused and reviewable.

Verification

• cargo test
• cargo clippy --all-targets -- -D warnings
• cargo fmt --check
• PHP playground syntax checks
• PHP playground runtime assertions
• Laravel demo syntax check
• Laravel analyzer reports exactly the three intentional errors

Checklist

• I have updated CHANGELOG.md
• I have updated the documentation where applicable
• I have added tests to cover my changes
• I fully understand the code that I am submitting, including code drafted with an LLM
• I have proof-read and copy-edited generated prose

🤖 Generated with Codex

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[shuvroroy]

CI note: the benchmark job reported a broad 31–76% slowdown across unrelated completion, hover, parsing, and diagnostics suites, then failed while trying to post its alert review because the fork token received a 403. The same filesystem-alias commit is included in #371 and #372, where both CPU and memory comparison jobs passed.

This looks like runner variance rather than a path-normalization regression. I cannot rerun an upstream workflow from the fork without repository admin permission; please rerun the failed benchmark job when convenient.

🤖 Generated with Codex

[AJenbo]

I'm on Linux so things might be a bit different, but I made some small corrections after testing:

The regression I fixed

view_names_for_blade_uri canonicalized the template path instead of trying it raw. A template that is itself a symlink into a shared directory then resolves out of its view root and gets no view name at all, so its callers' variables stop being typed. I added view_name_resolution_keeps_symlinked_templates, confirmed it fails on the commit as written (left: [], right: ["shop"]), and restructured the matching to try raw first and fall back to canonical only when the spellings don't line up.

That reordering is also a performance win over main: the old code canonicalized every view root on every call, and blade_render_order calls this once per Blade URI. The common case now costs zero syscalls, and the template path is canonicalized lazily through a OnceCell so it happens at most once even with several roots.