Spatial interpolator reducers

The default spatial interpolator applies the current value to the current position, resulting in a "last" binning reducer. We could implement others (thinking out loud for now):

• "first" is the easiest: just don't overwrite if a value is already present
• "sum" is easy too (though it applies only to quantitative values), with a caveat that the color domain will have to be set by hand as it will almost always be wrong (it's based on the input values sent to the interpolator); the contours should work directly
• "random sample" and "mean" can be done by keeping a separate track of the count; ("mean" expects the value to be quantitative)
• "median" (and any other reducers, including custom functions) need to keep track of all the values, so that's probably not going to be great performance-wise, unless we can have iterative reducers

No implementation files, tests, or entry points are named. Start by locating the spatial interpolator and define the supported reducer scope, including quantitative constraints, color-domain behavior, contours, and performance. Done means the reducer API and its behavior are specified and implemented with coverage for the selected cases.