fix(ext/node): support encoding option in fs.watch

[nathanwhitbot]

Summary

fs.watch(path, { encoding }) previously emitted the filename as a plain utf8 string regardless of the requested encoding. Match Node's lib/internal/fs/watchers.js:

• encoding: 'buffer' → emit a Buffer
• any other named encoding (e.g. 'hex', 'base64') → re-encode the filename via Buffer.from(filename).toString(encoding)
• default ('utf8' or absent) → leave the string unchanged

Enables parallel/test-fs-watch-encoding.js.

Test plan

• parallel/test-fs-watch-encoding.js passes via cargo test --test node_compat
• Existing fs-watch*/fs-promises-watch* tests still pass

[nathanwhitbot]

Encoding paths line up with Node's behavior — utf8/undefined pass-through, 'buffer' returns a Buffer, other named encodings via Buffer.from(filename).toString(encoding). Net effect is the same as Node's Buffer.from(<os-bytes>).toString(encoding): the original UTF-8 bytes get re-interpreted through the target encoding, including the same lossy behavior for utf16le/ucs2 on odd-byte filenames.

The watchListener type is also updated to string | Buffer, matching Node's typings.

[fibibot]

The substance is right and the helper is concise. Verified against the upstream test:

• encoding: 'hex' for filename 新建文夹件.txt: Buffer.from(filename).toString('hex') → 'e696b0e5bbbae69687e5a4b9e4bbb62e747874' (UTF-8 bytes hex-encoded), which is exactly the value the test asserts.
• No encoding (or 'utf8'): early-return the original string ✓.
• encoding: 'buffer': Buffer.from(filename) → real Buffer instance, satisfies the test's filename instanceof Buffer && filename.toString('utf8') === fn check ✓.

Buffer (line 92) and BufferEncoding (line 102) are already imported in fs.ts, so no new imports. watchListener type widening to (eventType, string | Buffer) => void correctly reflects the new return shape.

Test enrollment alphabetically positioned correctly (fs-utimes < fs-watch-encoding < fs-watch-file-enoent-after-deletion).

Holding off on autonomous APPROVE per fibibot's CI gate — only license/cla and lint title have run so far. The full CI build needs maintainer authorization to run on this bot's PRs. I'll re-confirm once test node_compat debug and test unit_node debug land green.

One small nit: passing an unknown encoding (e.g., { encoding: 'foo' }) currently falls through to asBuffer.toString('foo' as BufferEncoding) which Buffer will throw on at call time. Node validates the encoding upfront and throws ERR_INVALID_ARG_VALUE synchronously. Not a functional difference for the test, but validateEncoding would surface a cleaner error. Non-blocking.

[bartlomieju]

@nathanwhitbot this needs a rebase

[nathanwhitbot]

test node_compat (1/3) debug macos-x86_64 failed on test-process-threadCpuUsage-worker-threads.js — same macOS x86_64 timing flake we've seen on #33559 / #33601. That test isn't part of this PR's changes (it's already enabled in main). cc @nathanwhit