Indicate num_ffmpeg_threads is only relevant on CPU

[NicolasHug]

I could be wrong!
But if I'm not, I think we should just indicate that it's ignored on CPU. I don't think it's worth raising an error if users pass num_ffmpeg_threads=3, device="cuda" because it makes it really annoying to run "generic" loop and comparisons, e.g. stuff like

for device in ("cuda", "cpu"):
    decoder = VideoDecoder(..., num_ffmpeg_threads=1, device=device)  # no point errorring here

[scotts]

Seems correct to me!

[ahmadsharif1]

I don't think this is ignored for CUDA.

https://github.com/pytorch/torchcodec/blob/43ee8076e861a975a93fd019a95bf1860f827edf/src/torchcodec/decoders/_core/VideoDecoder.cpp#L438

It would be nice to run some experiments to see how it affects CUDA performance. And if we can confirm it doesn't affect performance, we can gate setting the threads on device=="cpu".

[NicolasHug]

Thanks for your input @ahmadsharif1 , I'll close this and we can follow-up on #353 then