Split schema file (w/lazy stats loading)

[radim]

For production databases schema.json is still much larger than what can be reasonable processed (even with summary level info).

Goal: split the files into schema.json and schema-stats.json, where only schema.json will be made (and possibly summary generated on import)