88 fix docstrings

[pat-alt]

No description provided.

[pat-alt]

@Rockdeldiablo to see. This PR

• improves the docstrings as discussed
• adds a CHANGELOG.md
• removes the redundant method overload

With respect to your ongoing work it is worth considering if what we need for conformal Bayes is just the GLM prediction (predicted mean and variance). In that case, I think our MLJFlux interface should use glm_predictive_distribution by default.

In the regression case, this is what is returned anyway when calling predict. In the classification case, the predict method just applies probit approximation (unless :plugin was specified) and optionally passes linear predictions through sigmoid/softmax.

[pasq-cat]

ok, i think i have understood.