alexrockhill/pd-parser: Usability

This version fixes handles of 'n/a's in the behavioral events used to synchronize with the photodiode. This may occur when the task suffers an error or a user goes in and removes some trials that they don't want to sync or don't have photodiode data to sync.

Additionally, the reporting of errors was improved: the number of samples will be reported if it's less than 3 times the resync amount. This had previously caused errors that were greater than resync to be reported as "none found" although an event could clearly be seen in the diagnostic plots. Now, the amount different is reported which makes more sense.

In noisy photodiode channels, there was additionally a problem with crossing and recrossing the zscore threshold. This was fixed by removing all events within max_flip_i of a found event.

Lastly, events were not recovered properly due to the same repeated events issues and not looking for events smartly as improved before. This was changed to match the process of looking for events during the alignment (examine every event greater than zscore but not immediately following another event greater than zscore and exclude events max_flip_i away from any found photodiode events).