Cleaner handling of music metadata retrieval

[bclindner]

Current implementation functionally pulls the track info twice (once at queue time, the other at runtime). Chances are nonzero for a YouTube ratelimit if requests happen too often. Also, this introduces ~1-2 extra seconds of latency that should be unnecessary

YouTube API is the obvious choice but I think we could also get away with streaming the URLs returned by the info JSON directly with a little finagling