Science error in horizontal interpolation at northern boundaries

The horizontal interpolation routines do not behave if points lie outside the northern bound of the original data. This was reported in a [MOSRS ticket](https://code.metoffice.gov.uk/trac/utils/ticket/188) years ago, but seems the fix never made it in to the trunk.

The rose-stem tests test running an n512 dump at n48 resolution, here shows going from an n48 dump to n512 resolution which demonstrates the problem:

We added the fix mentioned in the above Trac ticket in our fork (see [this commit](https://github.com/ACCESS-NRI/shumlib/commit/6adf418acdd62322f40e6ab31a9de36c59b24bfb)) and achieved desired results:

Thanks @MartinDix and @paocorrales for doing the work identifying this.

Start with the horizontal interpolation routines and the rose-stem tests that exercise an n48 dump at n512 resolution. Compare the existing implementation with the linked ACCESS-NRI commit, then run the interpolation test case to verify that points beyond the northern boundary produce the expected results.

Written by the indexing model from the issue text.