(Feature request) Generalized Maximum Likelihood Estimator for GEV

[jeremyfifty9]

A method for optimizing GEV parameters following the generalized maximum likelihood estimator method described in Martins and Stedinger (2000) would be very convenient for geophysical applications. In essence the GMLE method employs the maximum likelihood estimator with the addition of a prior on the GEV shape parameter. The prior is the Beta distribution with $\alpha=6$, $\beta=9$ in domain $[-0.5, +0.5]$ (but $\alpha$ and $\beta$ should be configurable).

As explained in the Martins and Stedinger (2000) paper, this is implemented by adding the objective function for the GEV with the objective function on the optimized shape parameter. For example:

shapepriorobj(ξ) = abs(ξ) <= 0.5 ? logpdf(Beta(β, α), ξ+0.5) : -1e16
gevobj(θ)        = loglike(model, θ)
nllobj(θ)        = -gevobj(θ) - shapepriorobj(θ[3])

where model is the EVA, $\theta$ are the GEV parameters, $\alpha$ and $\beta$ are parameters for the Beta distribution (with switched parameter positions because Distributions.jl and the Martins and Stedinger paper use opposite sign conventions). Additional checks may be necessary if the shape falls outside of the distribution range.