Feature: Percentage-based similarity

Right now, `looks-same` measures each individual pixel, and if even one exceeds the difference tolerance, it returns a failure. It would be nice if there was a mode where we could specify "I expect X% of the pixels to be the same".

In other words, if one or two pixels don't match, then still pass the test, but if a large number of pixels don't match, then it should fail the comparison.