(Rather than having, as before, the permutation merged into the Cg-P
step.) I'll revert this commit in a moment, to make sure the issue
really was just with the matrix parsing.
Turns out I had the wrong problem statement. I'm given input in CG
notation, but it is not enough to check under permutations for this
representation. Therefore, I should instead first convert to joint
probability representation before checking the permutations, which then
become trivial.
Treating the matrix entries as strings only served to introduce a level
of indirection which wasn't needed, since all of the given real-world
data has simple (very small) signed integer values. Now I just parse
those into a contiguous region of memory.