Dolt serializes and deserializes JSON unnecessarily.

[nicktobey]

Currently, when running a query on a JSON column, the following sequence of events happens:

• tree.GetField loads the entire on-disk representation of the JSON document into memory by calling tree.JSONDoc.byte(). This representation is currently a JSON string.
• tree.GetField converts this string into a types.JSONDocument by de-serializing the string into a golang map.
• Before being displayed to the user or send over the wire by the server, it must be re-serialized back into a string.

If the document is not inspected or modified by the query, we are deserializing and reserializing the document unnecessarily. I have benchmarks where this is 30% of the runtime for a simple select query.

We should have an optimization that detects when the structure of a JSON column is not actually needed for resolving a query, and treats the document as a text string instead.