Don't pass dims to the likelihood distribution

[tomicapretto]

We are passing dims to the likelihood distribution. Internally, there's something like what follows happening

with model:
    ...
    pm.Distribution(name, **params, observed=observed, dims=dims)

However, the dims aren't used and they can bring more problems in special cases. PyMC can determine the shape of the distribution from the shape of the observed data.

For more context, I figured this out because of a collaboration with people developing a library on top of Bambi. They have a custom likelihood function where observed.ndim is 2, but Bambi was passing a single dimension. My first attempt was to add a function that added as many dimensions as needed (a4e3646) but it turns we can actually make it work without passing dimensions and just letting PyMC use the shape from observed.

[codecov-commenter]

Codecov Report

Merging #629 (27bd9ea) into main (9ba92e1) will decrease coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #629      +/-   ##
==========================================
- Coverage   86.63%   86.63%   -0.01%     
==========================================
  Files          37       37              
  Lines        2200     2199       -1     
==========================================
- Hits         1906     1905       -1     
  Misses        294      294              

Impacted Files	Coverage Δ
bambi/backend/terms.py	97.36% <100.00%> (-0.02%)	⬇️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more