[lohedges]

Closes #270.

When perturbed_system is used to seed replicas, its box is copied to the replicas closest to the perturbed end state. Until now the only check against the reference system was the atom count. If the two end states were equilibrated with NPT, their boxes differ, so for GCMC in the NVT ensemble the replicas would start with different volumes.

This PR raises an error in _check_inputs when all of the following hold:

• replica exchange is on
• a perturbed_system is given
• gcmc is enabled
• pressure is None
• the two boxes differ

Boxes are compared by type and by their vectors (dimensions() for PeriodicBox, vector0/1/2() for TriclinicBox) with a relative tolerance of 1e-5, so a box with the same volume but a different shape is also caught. The error message suggests rescaling the perturbed system to the reference box, or setting a pressure to run in the osmotic ensemble.

GCMC in the osmotic ensemble (with a pressure set) is unaffected.

Tests

test_perturbed_system_box_gcmc covers four cases:

Pressure	Perturbed box	Expected
None	same box	passes
None	larger box	raises
None	same volume, different shape	raises
1 atm	larger box	passes

The check runs before any GPU setup, so the test runs on CPU. It's skipped on macOS, where GCMC isn't supported.