[c++] Fix offsets for nullable columns

[nguyenv]

Issue and/or context:

[sc-62104]

We were adding the offset to the ArrowArray's validity buffer directly which is incorrect because the validity is stored as a bit, not as a uint8_t.

Changes:

• Add util::bitmap_to_uint8 which takes a pointer to a bitmap, length, and offset, and returns a vector<uint8_t> representation; if the bitmap is a nullptr, then return a nullopt
• Add helper functions ManagedQuery::_cast_bool_data, formally util::cast_bit_to_uint8, and ManagedQuery::_cast_validity_buffer, which takes the validity buffer of the passed in ArrowArray and returns vector<uint8_t> with the properly shifted offset
• Update ManagedQuery::setup_write_column and ColumnBuffer::set_data to take in an std::optional<std::vector<uint8_t>> where the validity buffer has already been offset and casted to uint8
• Add pytest unit testing for nullable columns with offsetting for string, Boolean, enumeration, and the general case (int, float, etc.)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.28%. Comparing base (f3ba4e5) to head (b5e818c).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3611      +/-   ##
==========================================
+ Coverage   86.26%   86.28%   +0.01%     
==========================================
  Files          55       55              
  Lines        6378     6378              
==========================================
+ Hits         5502     5503       +1     
+ Misses        876      875       -1     

Flag	Coverage Δ
python	86.28% <ø> (+0.01%)	⬆️

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

Components	Coverage Δ
python_api	86.28% <ø> (+0.01%)	⬆️
libtiledbsoma	∅ <ø> (∅)

[bkmartinjr]

It would be good to add some additional tests:

1. Call Sparse/Dense NDArray write with an array that has a validity buffer, and ensure it acts as expected.
2. For DataFrame, need to verify expected behavior if an index (dimension) column is written with a validity buffer.
3. In test_dataframe, I'd also test the "invalid" timestamp types (NaT) alongside the other nullable types.

[bkmartinjr]

testing requests only at this point - the C++ code looks OK (I did a quick pass only - would be good to get @XanthosXanthopoulos or @jp-dark to review it in more depth)

[jp-dark]

I'm not sure the behavior for dense arrays is correct, but otherwise it looks good to me.

[nguyenv]

Updated now with all requested changes except:

Call Dense NDArray write with an array that has a validity buffer

Because DenseNDArray takes in a pa.Tensor which does not support nulls.

[bkmartinjr]

As I understand, soma_data also can't be nullable - can we add a branch to test for both?

[bkmartinjr]

One small (optional) suggestion. Otherwise LGTM!