Replace typesystem with Spark-/Hive- types

[okennedy]

Possibly required for #361
Would subsume #333

Mimir has a much cruder internal type system than Spark. In addition to lacking collection types, there's a lot of capabilities (e.g., integer precision) that Mimir just does not support at the moment. Spark already has a well-developed type system, and it would be fantastic if we could replace this part of Mimir with the corresponding parts from Spark.

• Replace Mimir's Type with Spark's DataType
• Update or replace Typechecker with corresponding parts of Spark (not sure where this lives... perhaps the Catalyst analyzer
• Tack on support for unit-based types (Maybe use User-Defined Types?)
• Move to serialization using foo.json and foo.as[DataType] (make sure to keep track of everywhere this happens)
• Add an upgrade path to migrate old, encoded types to the new form