Improve logic for deciding which sublattices are equivalent when checking ordering

[bocklund]

Currently, sublattices in ordered (partitioned) phases are considered equivalent by symmetry if their site ratios are equal. This is a poor heuristic and will fail to determine that sublattices 0 and 1 are equivalent in an FCC_L12 phase with sublattice model [['A', 'B'], ['A', 'B'], ['VA']] with site ratios [0.25, 0.75, 3].

Relevant for this fix will be the current heuristic in pycalphad models pycalphad/pycalphad#311, and the changes that will occur as a result of fixing pycalphad/pycalphad#345.