When the backend changes there's a drop in the playback. When it's in
PlayingState it could time how long it takes until the playback resumes and add
that time to the seek.

When the backend changes the paths are not restored correctly and it crashes

When a new AudioOutput is created with the same name and category it'll still cause
a notification if a device fallback was necessary. One such notification per
name/category or even ignoring the name sure is enough.