[c++] Bug with sliced Arrow-table writes

[johnkerl]

Describe the bug

The customer repro involved tiledbsoma.io.update_obs with:

• Millions of rows
• Dozens of columns
• A large but non-categorical string column, taken from adata.obs.index
  • As discussed on [python] Fix perf. hit for large enums on ingest from AnnData #3353, the AnnData write_h5ad / read_h5ad path tends to categoricalize string columns, which is why only the adata.obs.index column was affected

Collectively these resulted in an Arrow-table slice here:
https://github.com/single-cell-data/TileDB-SOMA/blob/1.15.0rc4/apis/python/src/tiledbsoma/io/ingest.py#L1128-L1134
and the slices other than the first had, for this one column only, values from the first slice written. The Python-level slicing is correct, and the sliced Arrow table in Python is correct; the incorrect data is only visible (a) when debugging the array-writes in C++, or (b) on readback.

To Reproduce

A simple repro has been encapsulated in #3433 as a new unit-test case.

Tracking: [sc-60767]