UI probably should remove the "stream" control and put all channels of all streams into the matrix.

May just be able to spit out audio from AudioDecoder with a channel offset, or something, and get AudioMerger to sort it all out in Player::process_audio.

I guess streams can have different sample rates, so AudioContent can no longer have a content_audio_frame_rate.

May need a AudioDecoder::_audio_position per stream; move it down to FFmpegDecoder / SndfileDecoder so that they can choose how they do it.

It's not clear whether FFmpegContent should keep the AudioMapping in the streams or in itself.

-1
Add AudioStream parent containing frame rate, mapping, number of channels
[bb87d96]

-2
Use AudioStream in SndfileContent

-3
Add stream object to AudioDecoder::Audio signal.

-4
Remove selected audio stream from FFmpegDecoder.
FFmpegContent to work out audio channels by looping over its streams.
FFmpegContent to convert mappings from whole-content to per-stream and back.
FFmpegDecoder to decode all audio packets and emit stream.
SndfileDecoder to emit a stream too.
Remove content_audio_frame_rate() from AudioDecoder.
Player to store Resamplers based on stream object.
Remove stream selection UI and add stream details to the matrix.

-5
Make _audio_position a map keyed by AudioStreamPtr.

Merged from merge-streams in the commits leading up to a3ae559cb5b94ea0b42ebf7683a1232ccf0b74df.