Fix broadcastables inference for Alloc and an as_tensor_variable bug when all ndim=0 and all dimensions are broadcastable

[brandonwillard]

This PR closes #692.

It generalizes the broadcastables inference used by RandomVariable, which performs constant folding and shape simplification rewrites instead of using get_scalar_constant_value.

[codecov]

Codecov Report

Merging #694 (c67e71b) into main (4e10772) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #694      +/-   ##
==========================================
+ Coverage   77.59%   77.61%   +0.02%     
==========================================
  Files         152      152              
  Lines       46926    46915      -11     
  Branches    10885    10883       -2     
==========================================
+ Hits        36412    36413       +1     
+ Misses       7904     7895       -9     
+ Partials     2610     2607       -3     

Impacted Files	Coverage Δ
aesara/tensor/basic.py	86.17% <100.00%> (+0.58%)	⬆️
aesara/tensor/extra_ops.py	88.46% <100.00%> (ø)
aesara/tensor/random/op.py	98.57% <100.00%> (-0.07%)	⬇️
aesara/tensor/basic_opt.py	83.91% <0.00%> (+0.12%)	⬆️