Implement test for `anemoi-inference` data loader

[leifdenby]

TODO:

• upload the anemoi-inference sample dataset to ATOS @mpvginde - /ec/res4/scratch/cu0k/anemoi-inf-examples
• try reducing that dataset i size (sub 50MB would be ideal)
• upload to EWC object store
• implement test based on test data stored EWC
• once tests pass for all three loaders then
  • tag v0.1.0 of mlwp-data-specs and mlwp-data-loaders

[observingClouds]

/ec/res4/scratch/cu0k/anemoi-inf-examples got already wiped

[mpvginde]

@observingClouds, the anemoi-inference examples are now again available at /ec/res4/scratch/cu0k/anemoi-inf-examples

[leifdenby]

I've now uploaded these sample files to the EWC bucket with sample datasets, for example in https://object-store.os-api.cci2.ecmwf.int/mlwp-sample-datasets/anemoi-inference/unknown-revision/anemoi-inference-lam_2020020100.nc

I was thinking it might be sensible to try and prefix the path to somehow reflect what "revision" of anemoi-inference was used to create this dataset, what do you guys think given @observingClouds already found something else claiming to be an "anemoi inference" dataset in #7

If you think this sounds good could you let me know what revision this is from @mpvginde ?

I've drafted a test in #10

[observingClouds]

I agree that a revision indicator would be good, but version numbers like found in the metadata of an "anemoi-dataset" are based on experience in e.g. ecmwf/anemoi-datasets#122 (comment) not trust-worthy. So only the creator knows which anemoi-dataset release is compatible. So I guess, we need to use the software release tag, rather than some ill-defined metadata version indicator.