Always use syspath conversions (#3690 split up, part 2)

[wisp3rwind]

Follow-up to #4830:

I've had #3690 in the pipeline for a long time; and it has grown too big to look at in one piece. This is just the first commit rebased on master. It is still quite big, but only touches test code, so should be safe (assuming CI is green).

This is part 2, which mostly concerns plugins (except for a small fix to the artresizer module such that it's internally consistent).

[sampsyo]

Very very cool—these all seem right to me.

I'll use this space to again lament that Python 3 went in the direction it did for paths; I think the paths-are-str approach was worth trying but did not work out in the end, and now it's too late to change.

[wisp3rwind]

I'll use this space to again lament that Python 3 went in the direction it did for paths; I think the paths-are-str approach was worth trying but did not work out in the end, and now it's too late to change.

One more place where I think Rust is getting it right — by having an OS-dependent type instead of trying to abstract away too much.

On the other hand, I wonder whether there's really any need to support non-UTF-8 paths these days.

[sampsyo]

Yeah, the Rust approach makes a lot of sense to me. But you're right—if we were building beets from scratch today, it would be tempting to simply reject all files with non-UTF-8 names. Crashing seems perhaps too harsh, but (for example) simply logging a message and skipping any nonconforming file could work!