extension of the GLM interface

[tlienart]

Working with GLM.jl is actually much more of a pita than I'd have thought.

This PR allows:

• standard OLS
• binary regression with a variety of link functions (logit, probit, cauchit, cloglog)

It also in theory allows

• Poisson and NB count regression

but not tested, one the reason being that GLM doesn't deal well with matrices with categorical stuff... (see JuliaStats/GLM.jl#240) the workaround for now would be to assume that the data is passed as a dataframe and use their formula paradigm but this is annoying + forces use of data frame.

For things like multiclass classification using the logistic regression, a user would be better off using Flux (or sklearn's)

I'd suggest merging this now, and maybe trying to get some direct help from GLM devs to improve the interface, at the moment I'm wasting a lot of time trying to understand their package which is unfortunately very terse on examples and that doesn't seem very efficient.

I'll move to the Lasso stuff from Multivariatestats for now.