Revive the bbox circle of confusion calculation for PO

[zpelgrims]

Now that we're redistributing everything, I can see a lot of bad sampling in areas that are heavily aberrated, but where the thinlens model produces a flat field