Fix load balancing with FEM

[s-mayani]

This closes #485

The issue is that after load balancing, the FEM space used by the solver does not update its partitioning of the element indices among ranks according to the new domain decomposition, and the member variable resultField_m also needs to update its layout.

This issue was correctly identified by the AI tool Hugo in #487, but not properly fixed (fixed the symptom instead of the cause).

To fix this properly, a new updateLayout member function is introduced in the LagrangeSpace, which is called by the LoadBalancer in alpine if using the FEM solver, in the same way that the updateLayout of the fields works in the LoadBalancer. This means that every time load balancing is done and the domain decomposition changes, the LagrangeSpace takes the new domain decomposition into account and repartitions the elements too.

[s-mayani]

Draft PR: I am still waiting for some tests on Alps, which is under maintenance.

[s-mayani]

Checked with alps, produces correct results even when forcing a load balancing call after 5 time-steps, multi-GPU (2 and 4 GPUs checked).

[s-mayani]

Lumi not ok: 8 GPUs fine, 16 GPUs wrong results. Looking into it.