Implement 1-hour rolling window approach to min-binning

[Nate-Wessel]

Working toward an approach where the binning window rolls along, discarding observations >= one hour older than its leading edge.

This would serve to prevent data from being aggregated together over long spans because of

• discontinuities caused by the query parameters themselves
• gaps just because the data is sparse

[Nate-Wessel]

Still need to adjust the join of the bins back onto the data, as that's still a rolling join based on the bin ends.

[Nate-Wessel]

There's some inefficient code in here, but I'm pretty sure the logic per se is now sound.