The validation is great, and it would be nice to be able to use it outside the context of a request (e.g., JSON file validation). However, the public interface is somewhat awkward due to the included optimizations.
Would it be in the spirit of things to add a simplified interface for generic schema validation? Or does supporting such a feature go out of the scope of Huma?

A few details that could potentially be abstracted away:

• The Registry parameter is used only if the schema is a reference (or is a parent to one, recursively)
• The PathBuffer could potentially be made optional
• ValidateMode could have a default (or be ignored?)
• ValidateResult could implement Error() and be returned as an error

I feel like a lot of the extra details could be abstracted away without giving up the optimizations.
For example, a config/option struct could be used that provides reasonable defaults.

Another approach might be to create a wrapper function that simplifies use (or even just suggest that users create one themselves).