BUG: Fix dimension discovery of within array ragged cases

When `np.array(...)` finds a shape mismatch while inspecting
an array (or array-like), the current logic is to use only
part of those dimensions (as many as possible).

This case was buggy in the new code, and is now fixed.
In general, we may want to change the logic here, e.g. to
always strip all dimensions of that array (allowing an object
array of arrays), but there are some corner cases where the
current behaviour somewhat "works", so this fixes a bug while
gh-17224 and the issues around it remain independently.

0 files changed, 0 insertions, 0 deletions