Generalize `DeviceInterface` to include audio decoding

[scotts]

Currently, our DeviceInterface is implicitly a video device interface. Inside SingleStreamDecoder, we dispatch differently based on audio and video:

torchcodec/src/torchcodec/_core/SingleStreamDecoder.cpp

Lines 1322 to 1327 in 6d72f11

	if (streamInfo.avMediaType == AVMEDIA_TYPE_AUDIO) {
	convertAudioAVFrameToFrameOutputOnCPU(avFrame, frameOutput);
	} else {
	deviceInterface_->convertAVFrameToFrameOutput(
	avFrame, frameOutput, preAllocatedOutputTensor);
	}

Ideally, we'd like to turn that line into just:

deviceInterface_->convertAVFrameToFrameOutput(
        avFrame, frameOutput, preAllocatedOutputTensor);

That is, we handle audio and video the same. In order to do that, we need to somehow get SingleStreamDecoder::convertAudioAVFrameToFrameOutputOnCPU into a device interface. Design questions:

1. Should we extend CpuDeviceInterface to handle audio, in which case we'd dispatch inside of it?
2. Or should we keep CpuDeviceInterface to be just video only, and create a new kind of device interface that is just audio?

I think 1 might be the better option, but I'm not sure. I'm confident we'll only ever do audio decoding on the CPU.