Consider techniques for documenting/validating inputs/outputs of filters

[aaron-hopkinson]

A common issue with e.g. pandas-heavy code, is that a type hint of pd.DataFrame is not really sufficient to document the expected inputs/outputs of functions – often this information is covered by unit tests/docstrings. Packages such as Pandera provide a way of validating dataframe-like objects.

With the inclusion of tabular dataframe processing on the horizon, we should consider whether tools such as Pandera have value in anemoi-transform.

(Similarly, we should also consider whether we can do something similar for our gridded datasets)