read rave_dim, sample_rate and stereo where possible

[elgiano]

Some small changes to read settings from RAVE models when possible, so to avoid unnecessary cli arguments. The only other thing I would like to do in this direction is to read the minimum latent_size of a trained diffusion model, so that we could only provide length_mul, but I couldn't figure out how.

• rave_dims: read from model, remove argument Read rave_dims from RAVE model, like nn_tilde does (rave.decode_params[0]).

• generate.py: read sample_rate from rave by default if supported (rave v2) RAVE v2 models store their sample_rate as rave.sr. Since it doesn't make sense to choose a different one, read sample_rate from rave if possible, otherwise require the --sample_rate argument

• generate.py: split stereo only if rave is stereo RAVE stores rave.stereo = True if exported with --stereo. Check this value before splitting decoded audio in left and right, so that the model can work also with mono RAVE models.

• generate: more informative progress bar and messages Instead of a progress bar with 1 step, show the progress bar from sample, which shows diffusion steps. In my opinion it gives a more meaningful sense of progress through the task.

[moiseshorta]

Hey!

Thanks for this contribution, I'll test this and the other PR in the coming days and merge.

Looks quite useful.

Best, M