Fix hdf.DataSource and improve grid handling

[nbren12]

Earth-2 MIP Pull Request

Description

Closes #83.

• adds tests of hdf5 DataSource
• renames to hdf5.DataSource
• Makes a separate earth2mip.grid.LatLonGrid object for python API use and decouple the implementations from the configuration object schema.Grid.

Checklist

• [x] I am familiar with the Contributing Guidelines.
• [x] New or existing tests cover these changes.
• [x] The documentation is up to date with these changes.
• [x] The CHANGELOG.md is up to date with these changes.
• [x] An issue is linked to this pull request.

Dependencies

[nbren12]

/blossom-ci

[nbren12]

I tested some examples manually:

• inference_ensemble with this config:

  {
  "ensemble_members": 4,
  "noise_amplitude": 0.05,
  "simulation_length": 10,
  "weather_event": {
      "properties": {
          "name": "Globe",
          "start_time": "2018-01-01 00:00:00",
          "initial_condition_source": "cds"
      },
      "domains": [
          {
              "name": "global",
              "type": "Window",
              "diagnostics": [
                  {
                      "type": "raw",
                      "channels": [
                          "t2m"
          ]
                  }
              ]
          }
      ]
  },
  "output_path": "outputs/graphcast",
  "output_frequency": 1,
  "weather_model": "e2mip://dlwp",
  "seed": 12345,
  "use_cuda_graphs": false,
  "ensemble_batch_size": 1,
  "autocast_fp16": false,
  "perturbation_strategy": "correlated",
  "noise_reddening": 2.0
  }

• python3 examples/workflows/dlwp.py

[nbren12]

/blossom-ci

[nbren12]

/blossom-ci

[nbren12]

/blossom-ci