Reduce redundancy in forecast verification rules and scripts

Currently we use different rules and scripts to evaluate baseline forecasts and ML forecast runs. This creates redundancies and increases the danger of inconsistencies as quite a large fraction of the logic to read and verify is actually duplicated in the scripts.

Instead, we should refactor the code to work with either grib (ML forecast) or zarr (baseline) forecast inputs and consolidate the existing rules.

Start by locating the existing rules and scripts for baseline zarr inputs and ML grib inputs, then compare the duplicated reading and verification logic. Done means the consolidated rules handle both input types and preserve the existing baseline and ML forecast evaluations; the issue names no specific files or tests.

Written by the indexing model from the issue text.