Correct what the TransmitFile comments say about EOF and the socket mode

[EdmondDantes]

Comments only, in io_transmitfile_work_cb.

A probe of synchronous TransmitFile on Windows 11 26200 (NULL OVERLAPPED; a blocking socket and one set up as libuv's uv__tcp_set_socket does, same results):

• A count or start past EOF fails with WSAEINVAL. It does not stop at EOF and report success, as the comment said: it sent none of a 4 KiB file and all or part of a 2 MiB one. A file that shrinks during the call therefore fails the transfer with UV_EINVAL; it is not counted whole.
• The file pointer after the call stands 32 KiB in for any larger file, which is why counting from it (4102d7f) was wrong.
• libuv's socket is non-blocking (FIONBIO) and the call is still synchronous with a NULL OVERLAPPED.

The clamp from 1521d69 stays; the code is unchanged. Recorded in true-async/server dev/DECISIONS.md (PR true-async/server#369).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!