Extend LogisticRegression and IRLS to a GLM bundle

The IRLS function in the LogisticRegression bundle (and the PoissonRegression bundle) is a generally applicable optimization routine used for all types of generalized linear model (GLMs), and so should be kept somewhere central to be referenced by various bundles.

Examples of response distributions for which GLMs (and hence IRLS) could be used are:

• binomial (logistic)
• multinomial
• Poisson
• quasi-Poisson
• negative binomial
• gamma
• inverse Gaussian

Other common procedures (such as deviance, AIC, LL calcs) should also be moved to ML_Core.