Inconsistent format breaks deserialization

[justinkillen]

Hi,

There are some inconsistencies that make loading the data file difficult. In particular:

1. At the top level, there is a map keyed by machine family...except that it also includes licenses, which has a completely different data definition
2. Within a machine type spec element, the cores element is sometimes an integer value (e.g. n2d-highcpu-16 == 16) while other times, it is the string value "shared" (e.g. f1-micro)

Would there be any appetite here to make the data format more standardized? I'm thinking we could start with a json-schema definition.