Change model caching to support local pytest runs

[bfhealy]

Running pytest nmma/tests/*.py should ideally pass all tests locally in addition to Github Actions. Caching the svdmodels in the workflow file, e.g.

nmma/.github/workflows/continous_integration.yml

Line 80 in 0395d1e

- name: Restore SVD models from cache

currently leads to the analysis pytest failing locally, since local_only is set to True in nmma/tests/analysis.py.

The cached Bu2019lm model on the Potsdam enlil machine is not necessarily up to date with gitlab/zenodo (which are also currently inconsistent, see #317). There are a couple ways we could address this issue:

1. Point the workflow to gitlab or zenodo and download the model from there at the beginning of each workflow
2. Include Bu2019lm.pkl and a specific filter file in the nmma/tests/data directory, using that path for the analysis test's svdmodels argument

[sahiljhawar]

The cache is not stored on Potsdam machines. These are the original models. Caches are stored on Github servers (and now I cannot see any of the caches under the Actions tab). Caching was done in order to prevent timeouts (because of Zenodo). But yes, it will make sense to add the filter specifc model file in the data directory (I am not sure why didn't I do that previously).