Test loaders

The change itself is trivial, but as it has the potential to break multiple recipes I think there should be some sort of regression test for this. If we want to get this PR in quickly then maybe that could be a separate issue?

_Originally posted by @SGallagherMet in https://github.com/MetOffice/CSET/pull/1721#pullrequestreview-3270901422_

We currently don't have any testing on loaders, which means we don't have the protection against regression that we would like for them.

We should add regression tests here. The big question is how. While we could individually test each loader, that would be a lot of work that isn't likely to produce a large gain, as each loader is supposedly independent.

Do we just need a single config with all diagnostics enabled, that we verify against?

Start by reviewing the loader implementations and the recipes they support, then inspect how a configuration with all diagnostics enabled is processed. Compare whether one representative configuration provides useful regression coverage or whether loaders need separate cases. Done means regression tests cover the agreed loader behavior and fail when a loader breaks.

Written by the indexing model from the issue text.