feat(python): Respect index order in DataFrame.to_numpy also for non-numeric frames

[stinodego]

vstack followed by a transpose will result in an F-contiguous array. column_stack creates a C-contiguous array. So we can use one or the other depending on the order requested by the user.

Structured arrays are both C and F contiguous since they are 1-dimensional.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 81.39%. Comparing base (4375930) to head (8c24cd4).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #16390      +/-   ##
==========================================
- Coverage   81.40%   81.39%   -0.01%     
==========================================
  Files        1408     1409       +1     
  Lines      184222   184507     +285     
  Branches     2960     2960              
==========================================
+ Hits       149965   150183     +218     
- Misses      33742    33809      +67     
  Partials      515      515              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ritchie46]

Have you benchmarked those? The transpose is free as that only changes the strides.

[stinodego]

Have you benchmarked those? The transpose is free as that only changes the strides.

I have not benchmarked this - in this case it's a correctness issue:

The default order is Fortran and this path is unchanged (it uses vstack followed by transpose).

If the user explicitly asks for C order, we used to return a Fortran-contiguous array in this path. So that was incorrect or at least surprising. For this path, we now use the column_stack approach.