Still not sure whether it's better to calculate
rref[pivot_row][col] * (rref[row][pivot_col] / pivot_val) (option 1)
vs.
rref[pivot_row][col] * rref[row][pivot_col] / pivot_val (option 2)
(i.e. a * b / c vs. "a * (b/c))
in terms of floating point error.
But I think option 1 (current commit) is better, since the scale factor
(rref[row][pivot_col] / pivot_val) is always <= 1 here (I think).
