Support the expanded extract.ai saved-model contract

[ebhills]

Summary

WranglesXL issue #1206 expands the saved extract.ai worksheet contract. WranglesPY needs the paired runtime/compiler support so saved models created in Excel compile consistently and map their settings to the Responses API.

Required behavior

• Accept strict JSON and restricted JSON-compatible YAML in saved-model schema cells.
• Accept bracketed lists, pipe/comma-delimited lists, and object shorthand such as value: number | uom: string.
• Reject unsafe or ambiguous YAML features, duplicate keys, non-JSON values, excessive depth, and malformed schema content.
• Default blank Required to every named property.
• Default blank Additional Properties to false for objects with named properties.
• Make named nested properties non-null by default while retaining nullable top-level outputs.
• Accept saved-model ReasoningEffort values none and low, mapping them to the Responses API while preserving explicit call-level precedence.
• Preserve legacy saved-model behavior and field/example compatibility.
• Provide a user guide covering Excel saved models and recipe-owned definitions.

Acceptance criteria

• Compiler and Responses payload tests cover the expanded contract and defaults.
• Existing focused extract.ai tests remain green.
• Documentation distinguishes Excel shorthand from normal recipe YAML.