Should discontiguous audio be valid?

[bmcfee]

Since merging #1125 , we no longer have a strict requirement for the input to util.frame to be contiguous. A discontiguous buffer will be copied to a proper arrangement to support framing.

Relatedly, util.valid_audio requires that the input buffer be F-contiguous to pass as valid. This requirement was originally put in to ensure that framing can operate smoothly if the audio buffer is valid.

The question now is: should we relax the requirement in valid_audio?

Why does this matter: if we want to make a multi-dimensional STFT, it will need to frame a multi-channel input: (2, N) -> (2, NFFT, F). However, soundfile returns C-contiguous buffers, and the whole thing breaks. If we relax the requirement in valid_audio, then the whole thing can proceed as expected, if not quite as efficiently as possible.

Related issue: #1130

[bmcfee]

As implied by the PR, I've convinced myself that this requirement is no longer necessary and should be dropped ASAP.