lfric2um supports only bilinear regridding

### Are there any linked Issues or Pull Requests?

_No response_

### Brief description

This issue covers a couple of different problems so needs breaking up into at least two pieces of work. But it's worth initially describing them together.

The `um2lfric_regrid_weights` module and `lfric2um_regrid_weights module` are not aligned as the former supports land and sea point fields, and permits a choice of bilinear or nearest neighbour interpolation for a given STASH code whereas the latter applies bilinear to everything.

Nearest neighbour regridding is selected by listing fields that need nearest neighbour interpolation in the `specify_nearest_neighbour` array of the `regrid_options` namelist. See the following for a starting list of likely items that need nearest neighbour interpolation.

https://github.com/MetOffice/lfric_apps/blob/main/rose-stem/app/um2lfric/opt/rose-app-nwp_gal9.conf

For bilinear interpolation to work on land- or sea- fields, one needs to account for masked fields otherwise the coastal points are badly skewed by including RMDI values in the interpolation. An example of a field that requires bilinear interpolation in um2lfric is STASH code 9, soil moisture.

ESMF appears to support methods for such interpolation without leaving holes in the dataset, but additional weights files need to be created and used in lfric2um.

### Further details of the issue.

_No response_