Add test for MEPS data example

[leifdenby]

Aim: Create a test setup where we ensure that npy-file based datasets can be read into neural-lam and the training items contain tensors of the right shape.

TODO

• [ ] reduce number of variables, size of domain etc in Joel's MEPS data example so that the zip file is less than 500MB. Calling it meps_example_reduced
• [ ] create test-data zip file and upload to EWC (credentials from @leifdenby)
• [ ] implement test using pytorch to download and unpack testdata using pooch
• [ ] Implement testing of:
  • [ ] initiation of neural_lam.weather_dataset.WeatherDataset from downloaded data
  • [ ] check shapes of returned parts of training item
  • [ ] create new graph in tests for reduced dataset
  • [ ] feed single batch through model and check shape of output
• [ ] add github action to run tests during ci/cd

Links:

• uploading to EWC: https://github.com/mllam/mllam-docs/blob/main/hosting_source_datasets_on_ewc.md
• example of using meps_example for testing from Leif's branch: https://github.com/leifdenby/neural-lam/blob/mllam-dataloader/tests/test_mllam_dataset.py#L78
• leif's branch with dependencies handled with pdm: https://github.com/leifdenby/neural-lam/tree/maint/refactor-as-package