Handling unstructured regional ML data (BRIS and FastNetUK)

We should look at ways that we can support handling of BRIS regional ML data in CSET. There are two main changes for this data to be ingested in CSET.

a) Converting the data from an unstructured grid to a regular lat/lon. It would be good to create a generally agnostic callback that recognises if data is unstructured, and handle it appropriately (thinking LFRic native output before it gets restructured through the SLAM program). We have a few options for doing this in SciPy;

`griddata`, and the 3-4x faster `LinearNDInterpolator` as this does not recreate the target grid each time. We could make this faster by using multi-cpu and `subprocess`, but it will be an expensive operation regardless.

b) Modification of metadata, possibly using a local lookup dictionary, and creating a pressure dimension from the variable name.