Mesh variation with `geom_pertub_fact`

[chrisrichardson]

When perturbing the mesh geometry randomly, the behaviour is different in serial and parallel, which makes it hard to check for correctness or even consistency.

Can we vary the geometry in the same way in parallel?