[c++] Fix fast path for SparseNDArray nnz

[johnkerl]

Issue and/or context: [sc-58277]

Changes:

Notes for Reviewer:

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 83.97%. Comparing base (29b2648) to head (6eaba1e).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3229      +/-   ##
==========================================
+ Coverage   83.86%   83.97%   +0.10%     
==========================================
  Files          51       51              
  Lines        5505     5505              
==========================================
+ Hits         4617     4623       +6     
+ Misses        888      882       -6     

Flag	Coverage Δ
python	83.97% <ø> (+0.10%)	⬆️

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

Components	Coverage Δ
python_api	83.97% <ø> (+0.10%)	⬆️
libtiledbsoma	∅ <ø> (∅)

[bkmartinjr]

I understand why this change "fixes" the problem, but I'm uncomfortable with the design (in both the original PR and in this fix). It relies on a heuristic (names), and disallows lots of cases where the fast path would work fine (e.g., other non-variable-sized dimensions)

Possible solutions to consider:

1. directly test the dimension to determine if it variable-sized. Only then fall back (ie.., don't test name or hard-wire to only int64)
2. Or, given that the most common case, simply wrap the nonempty-domain fetch in a try/catch, and call the slow path only when you can't get it.

Either of these seem far less likely to bite us in the future as they are not heuristics based

[johnkerl]

@bkmartinjr #3231 if you're willing to take this as a follow-on

and disallows lots of cases where the fast path would work fine

You're correct

• The code before [c++] Fix bug in nnz of variant-indexed dataframes #2990 was broken and wrong in ways that I documented on [c++] Fix bug in nnz of variant-indexed dataframes #2990, and tested for
• The change I applied on [c++] Fix bug in nnz of variant-indexed dataframes #2990 was too strict -- it disallowed the fast path in certain cases, including -- importantly -- the one fixed by this PR
• But we can go further and allow more cases to take the fast path

I do want to do that as separate PRs -- this one and the to-be-created tracked on #3231 -- precisely because this PR "gets it right" for the most important cases which are (a) default-indexed dataframes; (b) all ND arrays always (which must have int64 dims). So the follow-on PR will enable us to take the fast path for non-default-indexed dataframes, and that will be nice.