Add additional inequality satisfaction convergence criterion

[jonmaddock]

Ensure that inequality constraints are satisfied at the solution point.

This requires ukaea/PyVMCON#16 to be merged first, to actually call any additional convergence criterion.

This change will only affect optimising regression tests without f-values (currently only large_tokamak_nof.IN.DAT and stellarator.IN.DAT.) However, the large tokamak case already has satisfied inequality constraints, so it doesn't change. The stellarator test takes over 400 iterations running locally, so I haven't bothered running it. Before this is merged, I'll find current evidence of this actually changing the solution to having all-satisfied inequality constraints.

[jonmaddock]

After checking again on an updated main, the stellarator regression test did converge after 18 iterations, but it had 2 violated inequality constraints. With this additional convergence criterion, it now does not converge.

I haven't been able to find a current example where this requirement (satisfaction of inequality constraints) results in a different, fully satisfied, converging solution, although I have experienced this with the large tokamak regression test in the past. Currently the large tokamak regression test converges with satisfied inequality constraints without this additional criterion anyway.

I believe this should be merged anyway, as a solution isn't feasible without this criterion being met.

[timothy-nunn]

LGTM. Please make an issue regarding re-enabling Stellarator regression test.

[jonmaddock]

Thanks Tim: added #3245 for the stellarator regression test re-introduction.