Need faster way to select neuropil cut.

[mjlm]

Selecting the neuropil has become slower now that we're typically getting a larger number of cuts.

Two non-exclusive options:

1. Manual: Update the neuropil scatter and traces while Tab-ing through the clusters to select a neuropil region. I did a test implementation and it's a little slow (but OK). Most of the time is taken by the debleaching (precisely, the call to robustfit within the debleaching function) and could be sped up by using simpler debleaching for this feature...
2. Automatic: My main criterion for assessing the quality of a neuropil fit is the amount of negative-going transients in the subtracted trace. We could calculate the trace for all potential neuropil ROIs and automatically pick the one with the smallest mean(trace(trace<0)). Problem: This might not be robust to selecting nonsensical choices so we might have to pre-select potential neuropil ROIs with another criterion.