[python] Ingest somacore classes

[ryan-williams]

Accompanying SOMA "core" changes: single-cell-data/SOMA#244

It can be awkward having some implementations in that repo:

• I'm experimenting with a dask kwarg to ExperimentAxisQuery.to_anndata; EAQ impl has lived in the core repo, to date, but arguably should not
• @bkmartinjr has been working on improved "fast CSR" algorithms, which were previously implemented and used in the SOMA core repo, but arguably belong in an implementation repo like this one.

Changes:
Move files and classes from single-cell-data/SOMA to this repo:

• _query.py:
  • Axis
  • AxisIndexer
  • AxisQueryResult
  • ExperimentAxisQuery
  • MatrixAxisQuery
  • JoinIDCache
• _fast_csr.py
• _eager_iter.py

See also #3345 -- this isn't a performance PR per se but it will enable an ExperimentAxisQuery performance-improvement follow-up to #3328

[sc-59595]

[codecov]

Codecov Report

Attention: Patch coverage is 87.70833% with 59 lines in your changes missing coverage. Please review.

Project coverage is 85.81%. Comparing base (8b6ab64) to head (3017ad8).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3307      +/-   ##
==========================================
+ Coverage   85.53%   85.81%   +0.28%     
==========================================
  Files          54       57       +3     
  Lines        5703     6169     +466     
==========================================
+ Hits         4878     5294     +416     
- Misses        825      875      +50     

Flag	Coverage Δ
python	85.81% <87.70%> (+0.28%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
python_api	85.81% <87.70%> (+0.28%)	⬆️
libtiledbsoma	∅ <ø> (∅)

---- 🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[johnkerl]

Looks like straightforward code motion (as intended!) -- looks good to me

Holding off on approval until we get the somacore PR merged & a release tagged

[bkmartinjr]

We could also remove all of this factory support for indexers. It was added to break the circular dependency when the C++ indexer was added to tiledbsoma. At this point, there is no reason the query implementation can't just use tiledbsoma.IntIndexer directly.

[ryan-williams]

I looked into this a bit but couldn't immediately figure it out; suggest we leave it for follow-on work.

[bkmartinjr]

works for me. I need to do a bunch of work in this code, so I can take it on

[bkmartinjr]

This PR looks good to me, with one optional item I noted in comments (clean-up opportunity in query.py). I think we should do that cleanup, but I'm fine if it arrives in a later PR.

[bkmartinjr]

We should also remove this class-specific threadpool, and use Experiment.context.threadpool (the common threadpool used elsewhere). The implementation in somacore predated the existence of the SOMA context threadpool and at this point is obsolete/wasteful.

[bkmartinjr]

As an aside, I can do the threadpool cleanup when I add the partitioned reader -- if that is helpful.

[ryan-williams]

I've just added 3017ad8 to this PR, which was my attempt to address this.

AFAICT, it is now safe to assume Experiment.context.threadpool is not None. Existing code+test implied otherwise, but I'm guessing that's out of date / no longer relevant. Can someone confirm?

[bkmartinjr]

Yes, I agree that the tests are out of date. There is code in the repo that assumes threadpool must not be None.

We should update the tests!