Stream ARCO ERA5 data

[NoraLoose]

This PR is in collaboration with @ScottEilerman. I started from #355, but somehow could not push to it directly. The PR implements streaming ERA5 data directly from the cloud so that users do not need to pre-download their ERA5 data.

Changes

• Introduced ERA5ARCODataset class.
• Added gcsfs as an optional dependency to support cloud streaming.
• Introduced a new pytest marker stream to isolate streaming tests:

  pytest -m stream --stream

This runs only streaming tests and excludes them otherwise. This allows faster development since streaming tests take longer.

• Added streaming tests to CI.
• Added ERA5 streaming to documentation.

Other changes:

• Reorganized internal logic for handling use_dask and read_zarr workflows.

• Cleaned up conftest.py by removing unnecessary request parameters from fixtures.

• Adjusted fixture start_time values in conftest.py to more intuitive dates (changed from January 31 to February 2) since source data does not include January.

• Tests added

• Passes pre-commit run --all-files

• Changes are documented in docs/releases.md

• New functions/methods are listed in docs/api.rst

• New functionality has documentation

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[ScottEilerman]

Pending confirmation that the diffs in the data look fine, this is ready to go! Thanks for helping get it cleaned up!

[NoraLoose]

@ScottEilerman I'm still getting the warning:

RequestsDependencyWarning: Unable to find acceptable character detection dependency (chardet or charset_normalizer).
  warnings.warn(

in the docs notebook when instantiating the SurfaceForcing object with streaming. We can leave it for a future PR to try to fix that.

[ScottEilerman]

@ScottEilerman I'm still getting the warning:

RequestsDependencyWarning: Unable to find acceptable character detection dependency (chardet or charset_normalizer).
  warnings.warn(

in the docs notebook when instantiating the SurfaceForcing object with streaming. We can leave it for a future PR to try to fix that.

What seems odd here is that chardet-normalizer is a dependency of requests, and at least in a fresh environment, seems to get installed via both conda and pip. Is it possible the environment where these doc notebooks are running is maybe in a weird state?

[NoraLoose]

Is it possible the environment where these doc notebooks are running is maybe in a weird state?

That was indeed the case! Thanks!

[NoraLoose]

This is good to merge.