Fix small open_virtual_dataset bugs

[ayushnag]

• Closes [BUG] incorrect typing on get_s3_filesystem has led open_virtual_mfdataset astray #906
• Fixes load=True error by updating optional dependency group
• Updated tests

---

📚 Documentation preview 📚: https://earthaccess--923.org.readthedocs.build/en/923/

[github-actions]

👈 Launch a binder notebook on this branch for commit 8f9d27e

I will automatically update this comment whenever this PR is modified

👈 Launch a binder notebook on this branch for commit d5ca8c7

👈 Launch a binder notebook on this branch for commit 63f36b6

👈 Launch a binder notebook on this branch for commit 3d3a279

👈 Launch a binder notebook on this branch for commit 529b5ca

👈 Launch a binder notebook on this branch for commit f39b151

[ayushnag]

Note that this PR changes the dmrpp tests to no longer verify the result of open_virtual_dataset since that is checked in virtualizarr. The tests here now check that the expected dmrpp files are found, parsed, and can be loaded.

[betolink]

Nice, sorry for the delay. Is this the last thing we need before adding example notebooks and releasing 0.13.0? cc @ayushnag

[betolink]

oops, looks like there are some failing integration tests. e.g.

tests/integration/test_virtualizarr.py::test_open_virtual_dataset[AVHRR_OI-NCEI-L4-GLOB-v2.1] 

I wonder if it's related to JPL's outage. I'll debug locally.

[ayushnag]

@betolink Thanks for reviewing and pushing the fixes! I believe zarr v3 doesn't break the parser, v3 is not yet compatible with kerchunk. However, kerchunk is needed for the load=True param so it is ok to pin to v2. However, we will have to solve this problem soon since virtualizarr will also soon require zarr v3. The alternatives are to use an icechunk in-memory store to load the data or use a Zarr store designed for reading manifests like the one suggested here.

[betolink]

Ahhh good to know! ok so let's get this merged and then work on the notebooks so we can have a virtualizarr release next week! @ayushnag