[FEATURE REQUEST] closed form estimates under H_a

[adw96]

This is a place to keep track of increasing the speed of radEmu for the case of

• only categorical predictors (eg 2 or more mutually exclusive categories; a generalization of a single binary covariate)
• fitting under the alternative (= full or unrestricted) model
• flexible constraint (eg single category, mean, pseudo-Huber median)

I believe @ailurophilia has the algebra for this mostly figured out and is going to begin on implementation, but may just hand over the algebra to a statdivlab RA.