Additional GP kernels

[sbfnk]

Description

This PR introduces additional GP kernels: Ornstein-Uhlenbeck and Matérn 5 / 2, for more modelling choices especially for forecasting.

Someone would have to check the maths is correct before this is merged.

Initial submission checklist

• My PR is based on a package issue and I have explicitly linked it. not fixing issue but an extension of functionality
• I have tested my changes locally (using devtools::test() and devtools::check()).
• I have added or updated unit tests where necessary.
• I have updated the documentation if required and rebuilt docs if yes (using devtools::document()).
• I have followed the established coding standards (and checked using lintr::lint_package()).
• I have added a news item linked to this PR.

After the initial Pull Request

• I have reviewed Checks for this PR and addressed any issues as far as I am able.

[seabbs]

Nice! Very exciting

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if b4c211c is merged into main:

• ❗🐌default: 33.1s -> 41.9s [+12.88%, +40.5%]
• ✔️no_delays: 41.9s -> 52s [-5.74%, +53.8%]
• ✔️random_walk: 9.84s -> 10.4s [-7.09%, +19.12%]
• ✔️stationary: 20.2s -> 21.2s [-8.88%, +19.11%]
• ✔️uncertain: 59.7s -> 1.25m [-4.26%, +55.7%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[sbfnk]

(reference for the maths is the approximate hilbert space gp paper https://arxiv.org/abs/2004.11408 on p.4 - perhaps something we should mention somewhere)

[seabbs]

We should cite and discuss this tutorial: https://avehtari.github.io/casestudies/Motorcycle/motorcycle_gpcourse.html#4_Heteroskedastic_GP_with_Hilbert_basis_functions

It also looks to contain a better way to set L and a way to directly build the diagonal matrix

[seabbs]

This looks good. I've just looked at the code side so far and will look at the maths side tomorrow.

I think as is is fine but the Aki tutorial suggests there are now faster/more direct ways to write this and potentially some better approaches to parameterise.

[seabbs]

I'm going to look at other parameterisations in another PR. I think this looks good for now. Maths makes sense to me.