Fix/scotch tests

This MR updates the existing parallel tests to force the use of METIS in the case that Scotch is also selected (which becomes the default). In some cases, Scotch would partition the mesh differently to METIS, which potentially leads to slightly different L2 and LInf errors being reported. The only case I observed this was in the bfs_tg_par test.

I have also included a three tests in MultiRegions which explicitly use Scotch (only active in the case that Scotch is selected, of course).