Handle POSIX short writes in FileCasStrategy chunk write loop

The POSIX chunk-write path in FileCasStrategy::WriteChunk only treated
write() == -1 as failure. A partial write (0 < Written < Size) was
silently ignored and the outer loop advanced its cursor by the requested
byte count, leaving a hole in the on-disk chunk.

write() is allowed to return fewer bytes than requested on a regular
file (e.g. signal interrupt mid-syscall, filesystem-specific limits).
It is also allowed to fail with EINTR before making any progress. Both
cases would produce a truncated shard file whose contents no longer
match its IoHash, surfacing later as a hash mismatch on read or as
silently corrupt data depending on the code path.

Loop until all bytes are written, retry on EINTR, and treat a zero
return with bytes still pending as an error.

0 files changed, 0 insertions, 0 deletions