Super basic audio decoding POC

[NicolasHug]

This is a very basic audio decoding POC.

Its purpose is to gauge how much of the codebase we would need to change in order to support audio decoding, and where.

This PR currently breaks all video decoding, and only partially works: I successfully get some audio that I can listen to, but something wrong with it (it's noisy and probably not sampled at the correct rate).

import torch
from torchcodec.decoders import VideoDecoder

decoder = VideoDecoder("./test/resources/nasa_13013.mp4", stream_index=1)
print(decoder.metadata)

audio_samples = torch.cat([frame.data for frame in decoder], dim=1)
print(audio_samples.shape)

import soundfile as sf
print(sf.available_formats())
np_samples = audio_samples.numpy().T

sf.write("out.mp3", np_samples, samplerate=8000)