fs: normalize trailing dot segments for rm

[AmarWaqar-TSKLI]

fs.rmSync() dispatches to binding.rmSync() (std::filesystem::remove_all) while
fs.rm() and fsPromises.rm() use the JS rimraf, so the two forms disagree whenever the
trailing path component is . or ... Fixture is a/b/c/d, options are
{ recursive: true, force: true }:

path	resolves to	fs.rmSync before	fsPromises.rm before	after (both)
a/b/../.	a	ok, leaves a	EINVAL, removes nothing	correct
a/b/..	a	ok, leaves a	ok, leaves a	correct
a/.	a	throws, code is ""	EINVAL, removes nothing	correct
a/b/c/.	a/b/c	throws, code is ""	EINVAL, removes nothing	correct
a/b/c/../..	a	ok, leaves a	ok, leaves a a/b	correct
a/b/c/d/../../..	a	ok, leaves a	ok, leaves a a/b a/b/c	correct

The last two rows are the ones worth attention: both forms report success and remove
different directories.

Why the async form fails silently

_rmchildren() builds child paths by concatenating onto the unresolved path. Tracing the
fs calls for rm('<root>/a/b/c/../..'):

rmdir(<root>/a/b/c/../..)         -> ENOTEMPTY
readdir(<root>/a/b/c/../..)       -> ok, ['b']
  ... recursion removes <root>/a/b/c/d, then <root>/a/b/c
rmdir(<root>/a/b/c/../../b)       -> ENOENT
rmdir(<root>/a/b/c/../..)         -> ENOENT
=> resolves successfully, leaves a/ and a/b/ behind

The walk removes a/b/c, which the literal path needs in order to resolve. Every later call
then fails ENOENT, and rimraf() treats ENOENT as "already gone", so the failure is
reported as success.

The change

Resolve a trailing dot segment at the three rm entry points so both forms operate on the
same path.

Only a trailing . or .. is rewritten. That is exactly where the two implementations
diverge, and keeping every other path byte for byte identical matters: normalizing
unconditionally also rewrote ./foo to foo, which changed the resource string the
permission model reports on denial and broke test-permission-fs-write. Narrowing to the
trailing component leaves paths that already behaved correctly completely untouched.

This is not a new behaviour so much as making the three documented input types agree:
URL paths are already correct in every row above, because the WHATWG parser resolves dot
segments before the path reaches fs. string and Buffer paths were not.

Buffer paths are round tripped through latin1 rather than utf8, because filenames are
arbitrary byte sequences and a utf8 round trip rewrites invalid sequences to U+FFFD, which
would remove a different path than the caller asked for:

utf8    round-trip: CORRUPTED       612ffffe2e2e2f62 -> 612fefbfbdefbfbd2e2e2f62
latin1  round-trip: byte-identical  612ffffe2e2e2f62 -> 612ffffe2e2e2f62

Test

test/parallel/test-fs-rm-dot-segments.js covers nine path shapes across string, Buffer
and URL input, plus a directory whose name is not valid UTF-8. Each case asserts both that
the two forms agree and that the surviving tree is the one POSIX path resolution implies, so
the test still fails if both forms are wrong in the same way. It was written before the fix
and fails on current main.

Locally: the new test passes on this branch and fails on v26.7.0; all 280
parallel/test-fs-* tests pass; all parallel/test-permission-* tests pass except
test-permission-drop-ffi, which fails on a clean checkout too because the FFI fixture
library is not built. make lint-js is clean for the touched files.

Notes

Supersedes #61968, which has been stale since February. That PR took the same general
approach and the discussion there informed this one, but the implementation here is fresh
rather than carried over, so I have not added a co-author trailer. Happy to add one if that
is preferred.

fs.rmSync() throwing an error with an empty code (rows 3 and 4) is a separate defect in
RmSync() in src/node_file.cc, where the std::error_code translation is a hardcoded list
of four values and everything else becomes UV_UNKNOWN. This change stops a/. from reaching
that path but does not fix the mapping. Filed separately as #65884.

Disclosure: I used an AI coding assistant on this change. The behaviour matrix, the call
trace and the round trip check above are reproducible scripts run locally against v26.7.0 and
against a checkout of main; I have read through the patch and the test and can speak to
both.

[github-actions]

Welcome to Node.js, and thank you for your first contribution!

Before review, please take a moment to read:

• the guide for first-time contributors
• the contribution and automation policies
• the pull request guide
• the AI use policy
• the Code of Conduct

Please make sure every commit is signed off. For a first pull request, GitHub Actions require collaborator approval and Jenkins CI must be started by a collaborator or triager, so an initial wait is normal.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.17%. Comparing base (8af7545) to head (09a4064).
⚠️ Report is 47 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #65883      +/-   ##
==========================================
- Coverage   90.18%   90.17%   -0.02%     
==========================================
  Files         771      771              
  Lines      264619   265204     +585     
  Branches    50231    50367     +136     
==========================================
+ Hits       238653   239145     +492     
- Misses      16961    17020      +59     
- Partials     9005     9039      +34     

Files with missing lines	Coverage Δ
lib/fs.js	97.31% <100.00%> (+<0.01%)	⬆️
lib/internal/fs/promises.js	90.98% <100.00%> (-0.12%)	⬇️
lib/internal/fs/utils.js	96.37% <100.00%> (+0.08%)	⬆️

... and 64 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[MikeMcC399]

This PR is failing CI tests.

See Pull requests > Step 6: Test

Before submitting your changes in a pull request, always run the full Node.js test suite.

with further details under BUILDING > Running tests.

[AmarWaqar-TSKLI]

Thanks for catching that. The failure was a bug in the new test itself rather than in the
change: it seeded a fixture tree and then asserted the directory was empty. Fixed and
force-pushed as 09a4064.

Locally the test now passes against a build of this branch and fails on v26.7.0 as intended.
All 280 parallel/test-fs-* pass, and parallel/test-permission-* passes apart from
test-permission-drop-ffi, which needs the FFI fixture library built and fails the same way
on a clean checkout here.