headFile/tailFile in filesystem server corrupts multi-byte UTF-8 characters at 1024-byte chunk boundaries

[cloud-id-2012]

Description

headFile() and tailFile() in src/filesystem/lib.js read the file in
fixed 1024-byte chunks and call .toString('utf-8') on each chunk
independently:

const chunk = Buffer.alloc(1024);
...
buffer += chunk.slice(0, result.bytesRead).toString('utf-8');

When a multi-byte UTF-8 character (e.g. CJK characters, which are 3 bytes
each) straddles a 1024-byte chunk boundary, the trailing incomplete bytes
in one chunk are decoded independently from the leading bytes in the next
chunk, producing mojibake (typically U+FFFD replacement characters) at
that position.

Reproduction

1. Create a UTF-8 text file containing Japanese (or other CJK) text long
   enough that some character boundary falls exactly on a multiple of
   1024 bytes.
2. Call head (or tail) with a line count that requires reading past
   that 1024-byte boundary.
3. Observe corrupted character(s) near the boundary in the returned text.
4. A full read of the same file (no head/tail) does not show the
   corruption, confirming the bug is specific to the chunked head/tail
   path.

Suggested fix

Decode the accumulated byte buffer incrementally using a streaming
decoder (e.g. Node's StringDecoder, which correctly buffers incomplete
trailing multi-byte sequences across chunk boundaries) instead of calling
.toString('utf-8') on each raw chunk independently.

Environment

• @modelcontextprotocol/server-filesystem (installed via npm, dist
  build dated 2026-07-22)
• macOS, Claude Desktop