Add closed-form posterior for Gamma-Exponential observation model

[larryshamalama]

Related to #48

This PR provides a closed-form posterior solution to an Exponential observation model with a Gamma prior.

Some part was directly inspired from other rewrites and tests, but it would be nice to ultimately get to #125

[larryshamalama]

Tests seem to be failing - not sure what needs to be updated

[zoj613]

@larryshamalama see pymc-devs/pytensor#441 Cc @maresb

[maresb]

I had a look, and it's a result of #1512 coupled with erroring on Aesara warnings. Specifically, the first lookup fails, and numpy.distutils gets imported. This results in the deprecation warning, which gets translated to an error in pytest.

In normal circumstances, it appears that this warning is muted whenever this import occurs via Aesara, although I don't know by what mechanism. Nevertheless it still trips the pytest warnings filter. Perhaps @brandonwillard has more context and a suggestion for how to tackle this?