Bug: Inconsistent use of shift in square structure atomic positions

There is a bug in `SIAB/data/structures.py` within the `square` function. The first two atomic positions don't include the shift while the last two do, which could lead to incorrect structure generation.

This issue was discussed by @kirk0830 in PR #107: https://github.com/kirk0830/ABACUS-ORBGEN/pull/107#discussion_r1833754588