Top Level List Support

[CalebCourier]

Summary

Adds the ability to generate structured data where the outermost layer is a List/Array. This can be helpful when using LLM's to generate synthetic data, when mimicking REST APIs, etc.

This PR also removes an exception that what raised during validation when the data was not a dictionary. We now let the process enter schema validation regardless of its type. If the type is wrong, this will be caught and result in a SkeletonReAsk. Addresses #594.

Usage

A list structure can be specified as such:

The Desired List Structure

[
  { "field": "hello, world!" }
]

Using Pydantic

class Foo(BaseModel):
    field: str
    
guard = Guard.from_pydantic(output_class=List[Foo])

Using RAIL

<rail version="0.1">
  <output type="list">
    <object>
        <string name="field" description="Any random string value" />
    </object>
  </output>
</rail>

TODO

• Unit Tests
  • Rail.output_type
  • JsonSchema.from_xml
  • JsonSchema.from_pydantic
  • validator_base.check_refrain
  • validator_base.filter_in_schema
• Integration Tests
  • from_pydantic with List[BaseModel]
  • from_rail with output.type = "list"
• Notebooks?
• Documentation?

[CalebCourier]

Needs cross version test fixes. Can pick up mid week.