Improve speed of spatially-delimited filters on push

[DocSavage]

If a data instance implements a spatially-delimited filter (e.g., ROI), dvid will iterate through all key-value pairs, check if it can be skipped, then pushes the data if it cannot be skipped. For small ROIs this leads to a lot of wasted time as it scans large volumes.

Ideally the spatially-delimited filter can supply the iteration through the key space. This would require data-specific support since it would depend on how space is encoded in that datatype's keys.