[C]Worthy OAE dataset example

[TomNicholas]

• Closes Aspirational use case: [C]Worthy mCDR OAE Atlas dataset #132
• Changes are documented in docs/releases.rst

A list of everything about this demo that is janky and should be fixed before merging, in order of most to least janky:

• Functions fail intermittently for unknown reasons, either with a HTTP error (which makes some sense) or saying they ran out of memory, which makes no sense
• Adding per-task retries might fix this (Use lithops RetryingFunctionExecutor to automatically retry tasks #575), however Lithops retries don't work properly, see Many S3 clients created when using RetryingFunctionExecutor lithops-cloud/lithops#1429
• Coordinates being demoted when writing to Icechunk #574
• Needs option to cache full file to be merged Allow passing config to default_object_store #564, see Option to cache file in open_virtual_dataset #625
• Even then still relies on the unreleased develop branch of VirtualiZarr
• Can't glob for filepaths in bucket, so we need Use obstore / obspec for globbing in open_virtual_mfdataset #569
• combine_by_coords didn't work because it triggered a reindex, and I don't know why
• Can't pickle numcodecs codecs numcodecs#744
• ManifestStore can't load scalars Generate a test case for ManifestStore with scalar arrays #530
• Have to rename paths to non-http URLs, because @maxrjones 's cache PR generates http URLs, but Icechunk can't store them yet (Add support for virtual chunks via Http/Https earth-mover/icechunk#526) - EDIT: for cases that don't require auth this should now work, but is untested
• I had to add the --provenance kwarg to my local docker build and I don't actually know if I need that
• I have to manually paste my lithops credentials into the .lithops_config because they seem not to be discovered when set as environment variables in the notebook
• The open_virtual_mfdataset function is not documented, but was added in Add open_virtual_mfdataset #349 (docs added in Docs on open_virtual_mfdataset and scaling #590)
• open_virtual_mfdataset parallel kwarg is different to the parallel kwarg for xr.open_mfdataset, because the generalization here should be merged upstream Use lithops to parallelize open_mfdataset pydata/xarray#9932 (EDIT: This one isn't important)

I have workarounds for basically all of them, but they should all be understood and fixed.