feat: Divergence-free and Curl-free Matrix Value Kernels

[adam-hartshorne]

As described in "Learning divergence-free and curl-free vector fields with matrix-valued kernels" by Macedo and Castro, 2010.

They outline how to define Matrix valued RBF functions which are divergence and curl free, for use in vector flow fields.

[daniel-dodd]

Hi @adam-hartshorne, happy to support you with a PR if this something you are interested in seeing. Note we currently do not have abstractions for multi-output kernel gram and cross-covariances - this might be something we'll need first. But keen to develop/add this soon (also aim to add some multi-output Gaussian process models to MOGPJax, dependent on this). Interested to hear your thoughts. :)