[c++] Fail earlier if n_buffers is not as expected

[johnkerl]

While working on #3100 for #3057 / [sc-55679] (parent #2406) I ran into a cryptic error from Arrow. It turns out that when the number of buffers doesn't match -- 2 for non-variable-length validity/data vs. 3 for variable-length validity/offsets/data -- we can get message like

tiledbsoma._exception.SOMAError: ArrowInvalid: Column 4: In chunk 0: Invalid: First or last binary offset out of bounds
At:
  pyarrow/error.pxi(92): pyarrow.lib.check_status

Now, the dev situation in which I was setting up data to generate such a situation is perhaps incorrect -- but -- we should fail sooner, and more transparently.

The status quo is that I turned on libtiledbsoma trace-logging & saw lines like

[2024-10-01 16:02:55.054] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] column type i name myint nbuf 2 2 nullable true
[2024-10-01 16:02:55.054] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] create array name='myint' use_count=4
[2024-10-01 16:02:55.054] [tiledbsoma] [Process: 5942] [Thread: 5942] [debug] [ArrowAdapter] release_schema for myint
[2024-10-01 16:02:55.054] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] release_schema schema->name
[2024-10-01 16:02:55.054] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] release_schema schema->format
[2024-10-01 16:02:55.054] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] release_schema done
[2024-10-01 16:02:55.054] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] column type f name myfloat nbuf 2 2 nullable true
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] create array name='myfloat' use_count=4
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [debug] [ArrowAdapter] release_schema for myfloat
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] release_schema schema->name
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] release_schema schema->format
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] release_schema done
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] column type U name newattr nbuf 2 3 nullable true <---- here
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [trace] [ArrowAdapter] create array name='newattr' use_count=4
[2024-10-01 16:02:55.055] [tiledbsoma] [Process: 5942] [Thread: 5942] [debug] [ArrowAdapter] release_schema for newattr

where the 2 3 was the smoking gun.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.09%. Comparing base (40e2178) to head (8fc2750).
Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3107      +/-   ##
==========================================
+ Coverage   88.95%   89.09%   +0.13%     
==========================================
  Files          45       45              
  Lines        4293     4293              
==========================================
+ Hits         3819     3825       +6     
+ Misses        474      468       -6     

Flag	Coverage Δ
python	89.09% <ø> (+0.13%)	⬆️

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

Components	Coverage Δ
python_api	89.09% <ø> (+0.13%)	⬆️
libtiledbsoma	∅ <ø> (∅)

[johnkerl]

Related to #3111