Refactor ORM

[furious-luke]

Separate the abstract syntax tree for the query (Abstract Query Tree?) and the strategy used to construct the SQL. Use a Query base class, and Select, Insert, Update, and Delete subclasses. These should be concerned with only how they perform their operations in combination with subqueries. Use a "Q" class for combining queries that will require ordering/CTE-like combinations.

Make more use of the migration schema, as it will be invaluable for data migrations.

Tasks remaining:

• Implement "delete" operations.
• Write tests for common API operations (nested insert).
• Filters.
• Adding to returning clauses.
• Table column reverses.
• Implement model level queries.
• Implement new migration system based on new schema.
• Separate inserts and return queries.
• Fix model update to use ID in provided model instance.
• Remove old SQL generation, update GQL queries.
• Update make migrations.

[furious-luke]

Instead of splitting inserts and selects, I'll substitute the insert queries over the table relationships. This limits the selections to only what was created, which I believe is what we really want. I should be able to add the substitute automatically when processing nested insert values in the query code.