Batch support

[vincentqb]

Many (all?) of the transformers do not currently support batches, and operate on only one non-batched tensor at a time. This can be fixed as part while setting a convention on the meaning of the indices across torchaudio.

Since signals don't necessarily have the same lengths unless pre-processing is done (i.e. padding, truncating, etc), we should wait on nested tensors.

[vgoklani]

why was this closed?

[vincentqb]

why was this closed?

Thanks for pointing this out: it should have remained open :)

[vincentqb]

In the meantime, a workaround is to convert (batch, channel, time)-waveform into (batch*channel, time)-waveform, apply the transform, and then convert back.

[vincentqb]

Fixed with #337