fix(filesystem): preserve UTF-8 across head/tail chunks

[CryoThrust]

What does this PR do?

Fixes #4666.

headFile and tailFile previously decoded each 1024-byte read independently. A UTF-8 character split across a chunk boundary could therefore become replacement characters.

This change keeps tail reads as raw bytes until the selected range is assembled, then decodes once, and uses Node's StringDecoder for forward head reads so incomplete trailing sequences are buffered across reads.

Verification

• git diff --check passes.
• Local test execution was attempted, but the installed Node binary cannot start because Homebrew's simdjson dylib is missing (dyld: Library not loaded).

[CryoThrust]

The updated branch now includes a regression test for a UTF-8 character split at the 1024-byte read boundary. Build and filesystem tests are green on the latest commit. Ready for review.