Change grid data model to use a hash

[josephwegner]

Right now we use a single-dimensional array to store all of the grid elements, and simply rely on the index of the item to determine where on the map that element exists. This is a simple structure, and works reasonably well, but it introduces quite a bit of complexity:

• Resizing the map has profound effects on the data model, requiring a full layout update. This is both expensive and error prone
• It is expensive in storage, network, and processing costs. Maps with lots of open space are filled with tons of nulls that we must iterate through regularly
• It locks future development. The constraints on this data model are pretty high, and block things like arbitrarily sized grids.

There are a number of options for better grid systems. I would suggest we implement some sort of coordinate based hash (x,y or x,y,z) and use that as a key in a JSON object.

[josephwegner]

As a note, some preliminary work has started here. I am using cube coordinates to calculate neighbors for fills: https://github.com/josephwegner/hexamagon/commit/91fd7f6ad31b9b9b311f346b38368825105e0774#diff-c21d4984f1216f95c91cdeb77d0880e4R147