Reduce the size of ensembles on disk

[alecramsay]

... especially but not only if we need to increase the number of plans collected.

There are 3 levels to this:

• Using a more compact index representations of precinct geoids and district ids that track back to the user-friendly versions
• Storing successive plans as deltas from the previous plan incrementally -- I can currently "pack" an ensemble in this way but only as a batch process after an ensemble is stored on disk which is how it currently needs to be stored in order to be read back in
• Zipping the resulting file -- i.e., store the unzipped file in a temp location and zip it

Then reverse the process when reading an ensemble from disk.

This would allow us to store much larger ensembles in GitHub w/o having to resort to LFS (which can be very tricky, in my experience) and reduce transfer times (back from the cluster).

[alecramsay]

The maximum number of precincts is CA with 25,607. The maximum number of districts is the PA state house with 203.