Add gufunc signature to RandomVariable's docstrings

[rlouf]

In this PR I add gufunc signatures to the RandomVariables docstrings. The notation is clear and concise; with knoweldge of broadcasting rules and the definition of size the signature makes shapes predictible.

Most distributions behave like Elemwise Ops, with the exception of:

• Multinomial (), (n) -> (n)
• Dirichlet (k) -> (k)
• MvNormal (n), (n,n) -> (n)
• Categorical (p) -> ()
• Choice (x) -> ()
• Permutation (x) -> (x)

[brandonwillard]

This work is part of #695. We need to create a single, consistent interface for this information and apply it to all the (effectively) block-wise Ops—which includes all the scalar Ops used by Elemwise (i.e. we need to replace all the nfunc_specs).

[codecov]

Codecov Report

Merging #1160 (ac3e073) into main (6232637) will increase coverage by 0.21%.
The diff coverage is 100.00%.

❗ Current head ac3e073 differs from pull request most recent head 4b3af51. Consider uploading reports for the commit 4b3af51 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1160      +/-   ##
==========================================
+ Coverage   79.14%   79.35%   +0.21%     
==========================================
  Files         173      161      -12     
  Lines       48528    48282     -246     
  Branches    10322    10962     +640     
==========================================
- Hits        38408    38316      -92     
+ Misses       7628     7454     -174     
- Partials     2492     2512      +20     

Impacted Files	Coverage Δ
aesara/tensor/random/basic.py	99.00% <100.00%> (ø)
aesara/sparse/type.py	72.89% <0.00%> (-8.69%)	⬇️
aesara/link/vm.py	90.26% <0.00%> (-2.30%)	⬇️
aesara/link/numba/dispatch/scalar.py	86.00% <0.00%> (-1.34%)	⬇️
aesara/compile/function/pfunc.py	81.40% <0.00%> (-1.01%)	⬇️
aesara/link/utils.py	61.12% <0.00%> (-0.72%)	⬇️
aesara/printing.py	49.76% <0.00%> (-0.41%)	⬇️
aesara/ifelse.py	51.00% <0.00%> (-0.29%)	⬇️
aesara/tensor/var.py	88.06% <0.00%> (-0.19%)	⬇️
aesara/tensor/basic.py	90.04% <0.00%> (-0.10%)	⬇️
... and 33 more

[rlouf]

Regardless of how we implement the interface we need to communicate how the current shape systems works to the users in the documentation, and I think gufunc signatures are the right tool for that.

[ricardoV94]

I think those two were wrong

[rlouf]

They were!

[rlouf]

Squashed the commits.

[rlouf]

Should we merge this? This documents the current behavior of RandomVariables and as such is useful even before #695 is merged.

[brandonwillard]

Given how many of these are scalars and, as a result, have trivial signatures, it's hard to say how useful those particular cases are.

Regardless, it doesn't hurt to add them, but we need to do something about the gufunc/RandomVariable comment before that.