ENH: add bfill functionality to numpy._core

[iquelli]

Relates to #25823.

• Implement backward fill functionality to fill NaN values using the next valid value along the specified axis.
• Add tests to ensure bfill works correctly for 1D and 2D arrays.
• Update documentation to include examples and usage guidelines for the new bfill method.

Backward fill is useful in many data preprocessing tasks where missing values need to be filled with
the next available data point.

Examples
--------
>>> a = np.array([None, 2])
>>> a.bfill(axis=0)
>>> a
array([2, 2])
>>> a = np.array([[None, 2], [1, 1]])
>>> a.bfill(axis=1)
>>> a
array([[1, 2], [1, 1]])
>>> a = np.array([[None, 2], [1, 1]])
>>> a.bfill()
>>> a
array([[2, 2], [1, 1]])

Co-authored-by: @goncalobarias

[ngoldbaum]

Probably not worth spending a bunch of time debugging before the discussion stabilizes on the mailing list, but the address sanitizer test failure is a heap buffer overflow in the C code added in this PR.

By the way, we have biweekly community meetings on zoom if you'd like to drop by to discuss this effort face-to-face.

[mattip]

I will close this for now, the discussion on the mailing list and the issue point to bottleneck as a better home for this. Please reopen or open a new PR (that passes sanitizer tests) if the consensus changes.