Refactor Multivariate distributions due to new meaning of size

[ricardoV94]

The next Aesara release will include aesara-devs/aesara#446 which altered the meaning of size for multivariate distributions, to be more consistent with that of univariate distributions.

Size will now specify the number of replications + dimensions implied by batched parameters. It is equivalent to shape[:-ndim.supp]. We will likely need to update our multivariate distributions and/or some tests (e.g. in test_distributions_random.py)

These changes should also be incorporated in #5437

[michaelosthege]

Important detail:
It's size == shape[:len(shape)-ndim_supp], NOT shape[:-ndim_supp]

That's because

>>> (1,2,3)[:-0]
()

[ricardoV94]

Important detail:
It's size == shape[:len(shape)-ndim_supp], NOT shape[:-ndim_supp]

That's because

>>> (1,2,3)[:-0]
()

Stupid -0 slicing heh :D

[twiecki]

@michaelosthege Do you have time to work on this?

[michaelosthege]

I will pick it up once the CI is green again on main. Don't want to solve two problems at the same time