[python] Support default-coord reads of corner-written dense ND arrays

[johnkerl]

Issue and/or context: In support of PR #1768 with issue #1769.

Needles to thread for update_uns:

The existing tiledbsoma.io.update_obs, tiledbsoma.io.update_var, tiledbsoma.io.add_X_layer methods all take as argument a tiledbsoma.Experiment which is already open for writing. Goal: do the same for tiledbsoma.io.update_uns.

Common use-case:

• Ingest data from AnnData to SOMA
• Outgest to AnnData; run ScanPy, curation, etc
• Write the updates back to SOMA

Technical factors:

• Whereas all the X, obsm, varm, obsp, varp are SparseNDArray, uns can contain DenseNDArray which are ingested from numpy.ndarray
• If the after-uns has numpy.ndarray that the before-uns didn't (new data on update), there is no problem
• If the after-uns has numpy.ndarray of the same dimensions as the before-uns, there is no problem
• Schema evolution does not support modification of dense domains
• We can try delete-and-re-add:
  • tiledb.Array.delete_array of the array accompanied by tiledb-group operations to delete the key/URI pair and re-add the key/URI pair
  • Note that for local-disk and S3, the new URI will be the same as the old URI; for cloud the new and old URIs will always be different URIs
  • Problem with that: we cannot delete-and-re-add at the same timestamp
  • If we do del collection[key] and collection[key] = new_dense_nd_array then we get tiledbsoma._exception.SOMAError: replacing key 'a' is unsupported which is a check that must remain (see links below)
• Also we cannot do: del collection[key], close the collection, and re-open at the same timestamp because of
  • https://github.com/single-cell-data/TileDB-SOMA/blob/1.5.0rc2/apis/python/src/tiledbsoma/_collection.py#L567-L570
  • [python] Stop using the flush hack when writing metadata or group contents #957
  • [python] Point-in-time child object access fails when using explicit write_timestamp #919
• We could have update_uns do two opens-for-write: one to remove any shape-changing dense arrays, and one shifted by a millisecond to do the rest of the work
  • However this would mean tiledbsoma.io.update_uns would take a URI as first argument, different from all the other updaters in tiledbsoma.io which take an Experiment opened for write as first argument

Good feedback from @isaiah and @seth: simply have tiledbsoma.io's uns-ingestion logic write bigger-than-necessary arrays for dense ND arrays -- as we already do for sparse.

This is compliant with the SOMA specification.

Changes: As above.

Notes for Reviewer:

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Files	Coverage Δ
apis/python/src/tiledbsoma/_dense_nd_array.py	95.83% <100.00%> (+0.37%)	⬆️
apis/python/src/tiledbsoma/_sparse_nd_array.py	92.25% <ø> (-0.15%)	⬇️
apis/python/src/tiledbsoma/_tdb_handles.py	96.27% <100.00%> (+0.20%)	⬆️
apis/python/src/tiledbsoma/_tiledb_array.py	84.76% <100.00%> (+1.26%)	⬆️

... and 87 files with indirect coverage changes

📢 Thoughts on this report? Let us know!.

[nguyenv]

Thanks!