ALE_Finite_Strain_Plasticity: fix MPI aborts from writing into ghosted vectors

[maieneuro]

Summary

Under MPI, the ALE_Finite_Strain_Plasticity example aborted during time
stepping with the deal.II assertion !has_ghost_elements(). The cause was
several places that wrote arithmetic results directly into
previous_deformation, which is a vector with ghost entries and therefore
read-only.

This PR fixes all such writes so the example builds cleanly and runs to
completion under MPI.

Changes

• Fix two places where we modify a ghosted vector (6c6caa3): initial
  fix for the ghosted-vector writes surfaced by the assertion.

• Fix remaining ghosted-vector writes (b135642): two further writes into
  previous_deformation remained:

  • mesh_motion: previous_deformation = current_increment; *= -1;
  • thermal: previous_deformation += current_increment;

  Both are now expressed as named NewtonStepSystem member functions
  (add_current_increment_to_previous_deformation() and
  set_previous_deformation_to_negative_current_increment()) that perform the
  arithmetic in fully-distributed temporary vectors and only assign the result
  back into the ghosted vector, which triggers the ghost-value communication.
  advance_time() reuses the first of these, removing a duplicated update block.

Verification

Builds cleanly and runs to completion (all 80 time steps) under
mpirun -np 2 with no ghost-element assertions.

[bangerth]

Thanks for making my commit nicer by introducing the helper functions.

@marcfehling Since I wrote one of the commits, would you care to look this over and merge?

[marcfehling]

Thank you all! The patch runs on my machine in parallel. I will go ahead and merge.