Support models with unconnected nodes removed from input (LAM)

[dietervdb-meteo]

This PR adds support for inference from checkpoints created with models where unconnected nodes have been dropped from the input grid, see anemoi-training #171. For the moment the functionality is restricted to inference taking an anemoi-dataset as input.

The indices of those gridpoints in the dataset that remain as input for the model are stored in the checkpoint as the supporting array grid_indices. The current implementation is such that if this supporting array is present in the checkpoint then the input is automatically reduced to those points with indices listed in grid_indices.

A small update of the boundary forcing is also needed, the mask used is now (the negation of) the output_mask that can be found in relevant checkpoints.

This addresses #87

Update: also added small fix to allow for automatic loading of datasets with more complicated configurations where sub_dictionary['dataset'] is not necessarily a string, e.g. of the type

dataset:
    - dataset:
        cutout:
          - dataset: ${hardware.paths.data}/${hardware.files.lam_dataset} 
          - dataset: ${hardware.paths.data}/${hardware.files.forcing_dataset}
        adjust: all
    - dataset:
        cutout:
          - dataset: ${hardware.paths.data}/${hardware.files.lam_z_dataset}
          - dataset: ${hardware.paths.data}/${hardware.files.forcing_dataset}
        adjust: all

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.03%. Comparing base (1187d78) to head (f8ed73e).

Additional details and impacted files

@@           Coverage Diff            @@
##           develop      #95   +/-   ##
========================================
  Coverage    98.03%   98.03%           
========================================
  Files            3        3           
  Lines           51       51           
========================================
  Hits            50       50           
  Misses           1        1           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.