Normalize correlations by number of channels

[calum-chamberlain]

Is your feature request related to a problem? Please describe.

Seismic data is frequently "gappy". At the moment EQcorrscan does a lot of checking to make sure that data that have long gaps are not used, and that correlations are not computed in these gaps. However, when there are gaps that are allowed then the correlation sum is effectively suppressed during these gaps, making detections less likely. The "noise" in the correlation sum is also likely elevated in these gaps because there is less destructive interference.

Describe the solution you'd like
EQcorrscan now does a good job of keeping track of where gaps are in the data, and uses this to ensure that correlations are zeroed in those gaps. We could also normalize the correlation sum by the true number of channels used. This would result in a -1 -- 1 range correlation sum and (hopefully) reduce the impact of gaps on detectability.

Additional context
This is related to a figure that @ebeauce pointed out separately.