vLLM server accepts null sampling params to request defaults

The vLLM server now treats null values in nullable parameters like temperature and top_p as absent, allowing clients to explicitly request server defaults. This change aligns the API with the OpenAI specification and existing json_value conventions.

• Extended schema validation to handle null values for nullable parameters.
• Added has_field() logic to skip nulls in field evaluation guards.
• Renamed internal reference from has_field to has_value.

This update enables clients to rely on server-side defaults by sending null, ensuring compatibility with standard OpenAI API behaviors.