Publishing to multiple archives has races that may lead to corrupt archives

[MonsieurNicolas]

Looking at GzipAndPutFilesWork, it looks like GzipFileWork is run in parallel on the same files (each archive tries to compress the same bucket file for example).

I suspect this may lead to strange publish failures and even corruptions: it's possible that a gzip process is spawned right before a different archives spawns a "put" of the same file, if this happens an empty/partial file may get uploaded (also, we use stdout redirection to the final name in "process manager", not a safer "redirect to temp file + rename").