Fix column->row cell vector mismatch in integrators

[CompRhys]

Summary

The tests of the function do not use the row vector cell so this suggests this is a bug that it's used in the integrators.

see that tests of function don't use the swap axis: https://github.com/Radical-AI/torch-sim/blob/d2cee6628ff3b4a8da7557772cc12a669a0c09b4/tests/test_transforms.py#L267

fixes #171

[CompRhys]

True error still not showing up with lazy import route, pyi needed to get the language server to work with the lazy imports. reset and moved lazy imports to #179

[CompRhys]

#17 is also related to this as if we solve that we can just drop the call to roll the positions back into the cell from the integrators.

[janosh]

my preference is to tackle that then and address #17 here instead of merging just a minor patch

[CompRhys]

I would still merge this but just wanted to raise that to bump the idea

[janosh]

i think i found 2 more bugs in transforms.py were the code was still written to work with column vector lattices while the rest of torch-sim now uses row vector convention as of #112

need some eyes on the changes in 6b82750 1e7290e from preferably two reviewers @orionarcher @abhijeetgangan @CompRhys

[CompRhys]

These weird values for the expected result seems self-evidently to suggest broken logic to me? These values must be projected onto the lattice vectors which isn't what I would expect the function to do

[CompRhys]

import numpy as np
from pymatgen.core import Lattice, Structure

lattice = Lattice(np.array([[2.0, 0.5, 0.0],
                   [0.0, 2.0, 0.0],
                   [0.0, 0.3, 2.0]]).T)

# Create structure with a single H atom at the given position
s = Structure(lattice, ["H"], [[2.5, 2.5, 2.5]], coords_are_cartesian=True)

# Get the wrapped (folded) position in cartesian coordinates
wrapped_pos = s.frac_coords[0] % 1  # fractional coords wrapped to [0,1)
wrapped_cart = lattice.get_cartesian_coords(wrapped_pos)

print(wrapped_cart)

[CompRhys]

this gives the same answer as the old test. I think that also suggests that there was an issue with the old function as I am giving the lattice as column vectors (due to transpose) to a function that expects row vectors.

[CompRhys]

@abhijeetgangan this okay to merge?

[janosh]

looks great! thanks @CompRhys 👍