Fix NaN issue in Zarr to NetCDF conversion for Gaussian grids

[jehangirawan]

Summary

Fixes #1113

This PR addresses the NaN issue when converting Zarr data to NetCDF for reduced Gaussian grids.

Problem

The original code assumed a regular lat/lon grid and used unstack() to create rectangular dimensions. For reduced Gaussian grids (where longitude points vary by latitude), this created a rectangular grid filled with NaNs for non-existent grid points.

Solution

• Added automatic grid type detection (regular vs Gaussian)
• Preserve Gaussian grids as unstructured grids with CF auxiliary coordinates
• Use ncells dimension with auxiliary lat/lon variables (CF-1.12 compliant)

Changes

• packages/evaluate/src/weathergen/evaluate/export_inference.py: Core conversion logic

Testing

uv run ./packages/evaluate/src/weathergen/evaluate/export_inference.py \
  --run-id <RUN_ID> --stream ERA5 --output-dir ./netcdf_out \
  --format netcdf --type prediction --samples 1

Results:

• No NaNs in output files
• All grid points preserved correctly
• CF-1.12 compliant output
• Compatible with xarray, CDO, Panoply

Before: Most grid points showed as NaN or _
After: All valid data points preserved

Note

This PR implements native Gaussian grid preservation as discussed in the issue.

Checklist before asking for review

• I have performed a self-review of my code
• My changes comply with basic sanity checks:
  • I have fixed formatting issues with ./scripts/actions.sh lint
  • I have run unit tests with ./scripts/actions.sh unit-test
  • I have documented my code and I have updated the docstrings.
  • I have added unit tests, if relevant
• I have tried my changes with data and code:
  • I have run the integration tests with ./scripts/actions.sh integration-test
  • (bigger changes) I have run a full training and I have written in the comment the run_id(s): launch-slurm.py --time 60
  • (bigger changes and experiments) I have shared a hegdedoc in the github issue with all the configurations and runs for this experiments
• I have informed and aligned with people impacted by my change:
  • for config changes: the MatterMost channels and/or a design doc
  • for changes of dependencies: the MatterMost software development channel