Fix bugs with constrained GeomOpt & FixSymmetry

[PascalSalzbrenner]

Summary

Bugs

I fixed three connected bugs causing high failure rates for constrained high-throughput geometry optimisations.

• In both FIRE & LBFGS, cell_positions were not updated after the cell was adjusted by constraints, which meant the next iteration started with stale positions. Now they are resynced after the adjustment.
• In LBFGS, the history vectors mixed two different coordinate frames because prev_positions/forces were stored in the un-adjusted coordinate frame. Now they are stored after the adjustment, consistently with the new positions.
• FixSymmetry's max_cumulative_strain clamp prevented full geometry optimisation. Speculatively, this was added to prevent strain blow-ups which the other bug fixes seem to have fixed. Removed max_cumulative_strain clamp & reference_cells which were used in the strain clamping. In my testing with heavily distorted structures, they all converge now.

Testing

• Added constrained optimisation tests for FixSymmetry
• Add history vectors test for LBFGS optimiser
• Updated FixSymmetry tests to remove references to max_cumulative_strain clamp & reference_cells

Checklist

Before a pull request can be merged, the following items must be checked:

• Doc strings have been added in the Google docstring format.
• Run ruff on your code.
• Tests have been added for any new functionality or bug fixes.

[CompRhys]

Thanks for the PR! I was out on holiday last week but will take a look in the next few days

[CompRhys]

@janosh Can you provide a repo for the issue motivating #447 as this removes a bunch of what I merged in there as a temp improvement. I confused #447 with #450. Please ignore