extreme value distributions module

As per @TonyLadson's request in greta-dev/greta#1 distributions for extreme value modelling would be useful. These would be best placed in a module (though admittedly the boundary between general distributions and extreme value distributions is blurry).

Specific requests include:

- [ ] GEV
- [ ] Gumbel
- [ ] Gumbel type B
- [ ] Pearson III
- [ ] Log Pearson III
- [ ] Generalised pareto

Like the already implemented distributions, these should all have a reference R implementation, where possible from a widely used package and preferably all the same package. `evd` looks like it has the best combination of consistency and scope, and appears to be widely used (judging by download stats from cranlogs).

This is not currently at the top of my to do list, but I would be really keen to support someone else in developing this module!

Start by reviewing the existing distributions in greta.distributions and the proposed module boundary. Compare the requested GEV, Gumbel, Gumbel type B, Pearson III, Log Pearson III, and generalised Pareto distributions with the reference implementations available in the R evd package. Done means the requested distributions are implemented consistently with the existing package and have matching reference R implementations where possible.

Written by the indexing model from the issue text.