Write multi-level datasets

[forman]

Added experimental function zappend.levels.write_levels() that generates datasets using the multi-level dataset format as specified by xcube.

It resembles the store.write_data(cube, "<name>.levels", ...) method provided by the xcube filesystem data stores ("file", "s3", "memory", etc.). The zappend version may be used for potentially very large datasets in terms of dimension sizes or for datasets with very large number of chunks. It is considerably slower than the xcube version (which basically uses xarray.to_zarr() for each resolution level), but should run robustly with stable memory consumption. The function requires xcube package to be installed.

Addresses #19.

Checklist (strike out non-applicable):

• Changes documented in CHANGES.md
• Related issue exists and is referred to in the PR description and CHANGES.md
• Added docstrings and API docs for any new/modified user-facing classes and functions
• Changes/features documented in docs/*
• Unit-tests adapted/added for changes/features
• Test coverage remains or increases (target 100%)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.93%. Comparing base (c5270c5) to head (b6af3ec).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #102   +/-   ##
=======================================
  Coverage   99.93%   99.93%           
=======================================
  Files          30       30           
  Lines        1458     1458           
=======================================
  Hits         1457     1457           
  Misses          1        1           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[AliceBalfanz]

I am missing the possibility to :

• define that I want to add new leveled time stamps from a certain point of a source cube to an existing multi level dataset. This is relevant for NRT services, where the base cube is "growing" and the levels need to be updated accordingly
• to write a multi level dataset with zappend from a xarray dataset, which does not exist on disk yet. I.e. I create a new dataset in memory and want to write it directly with zappend to a mulitlevel dataset. This is already possible for non-leveld data e.g. zappend(DatasetSlices(source_ds), force_new=False, config=zappend_config)

another case, but currently not urgent:

• concatinate two existing multi level datasets along e.g. the time dimension

[AliceBalfanz]

Works nicely! Thank you for the implementation!
I am missing two things in the documentation:

1. To inform the user, that write levels is not as robust as zappend itself. There will be a corrupt levels if an interruption happens during the process
2. Make the user aware of the need to remove bnds when leveling.

Do you think it is worth adding the above information? Not sure where the right place would be, maybe the changelog?

[forman]

Do you think it is worth adding the above information?

Yes!

[forman]

Not sure where the right place would be, maybe the changelog?

The function's docstring.