Add HTK format support to sox_io's save & info

[imaginary-person]

With advice & help from @mthrok, added htk format to sox_io's save function & info test.

htk is a single channel 16-bit PCM format used by HTK, a toolkit for building Hidden Markov Model speech processing tools.

[mthrok]

Hi @imaginary-person

Thanks for the contribution. The PR looks good. Can you resolve the conflict due to #1275 ?

[imaginary-person]

Can you resolve the conflict due to #1275 ?

Done. Thank you!

[mthrok]

Thanks!

[mthrok]

@imaginary-person

I forgot to mention adding one validation. Could you do a follow-up?
Since HTK only supports one-channel data, we would like to raise an error if the input tensor is multi-channel. In the following two locations can you add a check for the number of channels, similar to amr-nb case?

audio/torchaudio/csrc/sox/io.cpp

Lines 265 to 271 in 3488f31

	if (filetype == "amr-nb") {
	const auto num_channels = tensor.size(channels_first ? 0 : 1);
	if (num_channels != 1) {
	throw std::runtime_error(
	"amr-nb format only supports single channel audio.");
	}
	}

audio/torchaudio/csrc/sox/io.cpp

Lines 131 to 135 in 3488f31

	if (filetype == "amr-nb") {
	const auto num_channels = tensor.size(channels_first ? 0 : 1);
	TORCH_CHECK(
	num_channels == 1, "amr-nb format only supports single channel audio.");
	}

[imaginary-person]

@mthrok, thanks a lot for all your help so far!
I submitted #1291, but don't have the requisite permissions to add you as a reviewer.