Refactoring of SI3 routines in preparation for GPU offloading

Refactoring of SI3, specifically to move away from the gather-scatter approach that packs sea-ice points into 1-D arrays. This existing approach does not lend itself well to GPU optimisation, so the idea is to revert to working with the original 2-D (lat-lon) sea-ice arrays and apply a mask so that work is only done on points containing sea-ice. After implementing this method, it will be necessary to check that the CPU-only performance is not compromised before work begins on GPU offloading using PSyclone.

Initial tests will be performed using the BENCH configuration, then we will proceed to test against ORCA1 and ORCA12.