Schema constraints for OpenAI not supported

Description

Currently, LangExtract supports schema-constrained extraction with the use_schema_constraints option, but this feature is not available when using OpenAI models (via the openai backend) unless fence_output=True. However, OpenAI now provides official support for JSON schema, allowing structured output without relying on output fencing. Support for YAML formatting is not available via OpenAI's JSON schema mechanism.

Feature Request

Enable the use of use_schema_constraints=True with fence_output=False when leveraging OpenAI models, but restrict this functionality to cases where FormatType.JSON is used. If users select FormatType.YAML, the schema constraint option should remain unavailable or raise a clear exception.

Acceptance Criteria

• When using language_model_type for OpenAILanguageModel (e.g., "gpt-4o" or "o4-mini"), and FormatType.JSON, allow use_schema_constraints=Truewithfence_output=False`.
• When FormatType.YAML is specified with OpenAI models, do not enable schema constraints, and provide an explicit error message or fallback.
• Update documentation, tests and error messaging to clarify this limitation and the newly supported use case for OpenAI.

References/Related Work

• Existing Gemini output schema enforcement implementation.
• Possibility to leverage libraries like pydantic for runtime validation.