One job across servers, clearer failures, smaller databases

Fixes

• A preview job or schedule that picks libraries on more than one server is now one job: each shared file is processed once and published only to the servers you picked. Retries keep the same picks. (#426)
• A file FFmpeg can't open (missing, damaged or encrypted) now fails in one attempt instead of being retried on every GPU and then the CPU, and no GPU is blamed for it.
• A file that's replaced while it's being worked on is reported as replaced, not failed.
• Error messages show FFmpeg's real error instead of its progress line, for previews and loudness.
• The app log has one line per failed file. The step-by-step FFmpeg detail is in the job log, including from remote nodes.
• Intro & Credits job logs no longer write a line for every unchanged file (a nightly TV log went from about 135 MB to about 1 MB).
• Library Health's database gives back space it no longer uses, and finished loudness jobs no longer leave resume files behind. This frees hundreds of MB on large libraries.
• A Sonarr upgrade no longer logs a mount warning just because the old file was replaced.

Other

• Plain-language pass over the docs site and READMEs.
• The Emby and Jellyfin plugins have a catalog icon. They were released separately as Jellyfin plugin 10.11.1.2 and Emby plugin 1.0.0.2.