Fix weird peak properly

[rmflight]

So my solution to #32 was to just remove the offending peak and keep going.

However, we do have the point-point frequency distances already in the point metadata. After initial peak finding, it should be trivial to check them and remove any points that are outside the previously discovered frequency-frequency point distance, and then do the characterization.

Then if we still have an NA result, we still turf it. But at least we've checked (and the check shouldn't impact performance), and tried to make sure the data is correct going into the characterization calculation.

This means we rescue the peak, avoid the error in the first place if possible.