Fix stack last-axis layout mismatch - #21

This PR fixes NDArray::stack so stacked results use C row-major storage that matches the strides PHP assumes for indexing, getAt, and toArray.

Motivation and context

ndarray::stack could return a valid shape with a non-standard memory layout, while PHP always derives strides from shape alone. That mismatch made last-axis stacks look correct by shape but wrong in real use until data was read back element-wise.

What’s changed

• Fixed: stack (all dtypes) now normalizes output to standard row-major layout so logical indexing matches the underlying buffer.
• Added: regression test for stack(..., -1) on three 2D planes—shape, strides, contiguity, per-element equality vs inputs, and logical getAt checks.

Breaking changes

None.