Add filters to support creating tabular (observations) datasets

[aaron-hopkinson]

This issue covers creating filters in anemoi-transform to support observations datasets within anemoi. As a first step, we will look at porting across existing AI-DOP routines that would be appropriate to be used as filters in the context of anemoi-datasets.

To-do:

• Identify a list of processing functions which can be converted to filters.
• Add unit tests for all existing processing functions.
• Support the selection of "filters" via a registry (so that filters can be selected from either anemoi-transform or the existing code during the porting).
• Create base class for new filters, matching anemoi-transform interface.
• Convert filters to class based structure matching anemoi-transform filters.
• Ensure the filters have a consistent interface, i.e. accept a dictionary as a config argument only.
• Reorganise the file structure of the filters so they can be moved across to anemoi-transform in bulk.
  • (Still need to break out dependent functions so filters can be cleanly moved)
• Create integration tests with small (complete) configs.
  • @pinnstorm / @pwlean to add more configs?
• Make keyword arguments/names (i.e. yaml interface) consistent across filters.
• Update docstrings/documentation for new filters.
• Improve type hints/add schema documentation to all filters (basically "flavours" of dataframe).
  • Not a blocking feature, so created Consider techniques for documenting/validating inputs/outputs of filters #195 to cover this work.
• Move code across to anemoi-transform.
• Delete filters in existing repo.
• Throughout: ensure existing configs are updated to work with changes.