fix: handle enum types correctly in json_schema_to_model

[octo-patch]

Fixes #291

Problem

SendEmailNode (and other nodes with has_fixed_output=True and enum fields) throws a Pydantic validation error at runtime:

ValidationError: 1 validation error for SendEmailNode_1
provider
  Input should be a valid dictionary or instance of EmailProvider
  [type=model_type, input_value=<EmailProvider.RESEND: 'resend'>, input_type=EmailProvider]

Root cause: When setup() processes output_json_schema for a node with has_fixed_output=True, it calls json_schema_to_model(). That function recursively calls itself on every $defs entry — including enum definitions like EmailProvider. Because enum schemas have no properties, json_schema_to_model creates an empty Pydantic model for them. Later, when json_schema_to_pydantic_type resolves the $ref, the field type becomes that empty model instead of str. Pydantic then rejects the actual enum value with "Input should be a valid dictionary or instance of EmailProvider".

Solution

In json_schema_to_model, before recursively converting a $defs entry, check whether the definition contains an "enum" field. If it does, map it to the appropriate Python primitive type (str, int, float, bool) instead of creating a Pydantic model. This correctly handles all JSON Schema enum types that reference a primitive base type.

Testing

Manually traced the code path for SendEmailNodeOutput.model_json_schema():

• EmailProvider definition ({"enum": ["resend", "sendgrid"], "type": "string"}) now maps to str in defs
• provider field resolves to str, accepting both enum instances and raw string values
• Existing model/object definitions continue to be processed via json_schema_to_model as before