Update xarray usage in RAiDER to use lazy loading

[jlmaurer]

xarray makes available two methods for opening datasets, open_dataset and load_dataset. Using open_dataset allows for lazy loading and memory efficiency. This PR change all the locations were load_dataset was used to `open_dataset', resulting in much better memory performance for RAiDER. This PR also wraps netcdf dataset opening with try/except/finally logic that ensures the datasets are closed properly.

The PR also fixes a small issue where ECMWF requests were being handled in two API requests instead of one.

All unit tests pass locally.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)

Checklist:

• I have added an explanation of what your changes do and why you'd like us to include them.
• I have written new tests for your core changes, as applicable.
• I have successfully ran tests with your changes locally.
• My change requires a change to the documentation.
• I have updated the documentation accordingly.

[sssangha]

Hello @jlmaurer

I reviewed and tested the changes locally, and added two small safety enhancements:

1. Robust Resource Cleanup with ExitStack:
   To guarantee that all opened datasets are cleanly closed—even if an exception occurs mid-loop or during .to_netcdf()—I wrapped the file loading using contextlib.ExitStack:

   from contextlib import ExitStack
   
   with ExitStack() as stack:
       datasets = [stack.enter_context(xr.open_dataset(f)) for f in wfiles]
       # ... perform calculations and save to netcdf ...

2. Preventing In-Place Variable Reassignment
   Assigning ds_out = datasets[0] mutates datasets[0] directly while its disk handle is open. Updated this to explicit shallow copying:

   ds_out = datasets[0].copy(deep=False)
   

Future Consideration on Dask Chunking
Standard lazy loading via open_dataset handles the current memory performance improvements well. If processing ultra-large weather model grids ever hits RAM limits down the road, we can revisit introducing explicit chunking (e.g., chunks={}) to leverage Dask streaming directly to disk.

LGTM overall! I'm ready to approve if these changes look good to you.

[jlmaurer]

@sssangha thanks, looks good! I checked the PR again and found a couple small issues related to when I pulled it out from #810, so I fixed those (this is in ecmwf.py and related tests). If you want to just take a quick look at those, we should be good here.

[sssangha]

LGTM