Inlier stats (after checking for support) are incorrect

[johnwlambert]

Computation in L82 will give us the wrong numbers about inlier stats. If for 3 pairs, we had # inliers as 20, 20, 5, and if we threw away the 3rd, then we should have avg # inliers = 20.

But in our code it would be return as 15, which isn't correct.