Scores Aggregation: Adjust regrid.interpolate_to_point_cube for handling multiple forecasts and refactor

### Something else
As part of supporting cubes that have multiple forecasts (i.e. a `forecast_reference_time` dimcoord), we need to make some modifications to `regrid.interpolate_to_point_cube` to handle this. For example, extracting on the diagonal elements of the 2D cube assume a cube shape, which will no longer be appropriate for cubes with this additional forecast dimension.

We also need to refactor this, as it currently checks there are matching times in the forecast field cube and the point cube, yet we have operators in CSET already that check and retain common time points, such as `misc._extract_common_time_points`. If this operator is not sufficient for the needs here, we should adapt the misc operator so it is sufficient.

Some example code on branch `2409_rmse_agg_hinton` in solution for the `regrid.interpolate_to_point_cube` https://github.com/MetOffice/CSET/pull/2419/changes#diff-e0b1d7eb63d457f7b848b83b40a7eb67158ee93808258438b3dc1963de3f6161