Chunk a vector window read into sub windows

[mmcfarland]

The process of using the bounding box of a vector input to create a windowed read on a RasterioReader creates a situation where a MemoryError could be introduced if the cells in that window exceed the memory capacity of the machine. The bounding box window could be converted to several smaller window subsets, allowing a single machine to work on areas larger than it can store in memory.

A workflow might be:

• Calculate bounds for a given input vector
• Sub-divide that bounds into smaller units
• Iterate sub-bounds, and rasterize the full vector geometry on the portion to get a mask
• Compute statistics on just the masked sub-bound and store
• Aggregate the statistics for a final result

The caveat is that by reading smaller windows within the bounds, it limits the types of operations that could be performed since not all data is available.

Some form of this would need to be implemented to enable #9

[mmcfarland]

A good implementation of this idea can be found at https://www.azavea.com/blog/2016/10/05/philippines-road-safety-using-shapely-fiona-locate-high-risk-traffic-areas, "Optimizing intersection calculation".

That function could probably be used with very little modification to achieve the goals listed above.