Add support for `LanguageModelFeedback` and `logFeedbackAttachment`

The package should expose the above to make it simple for developers to submit feedback without resorting to rewriting failing inferences in Swift for the purpose of submitting feedback [1].

I have a suite of test cases running against FM, many of which are triggering erroneous guardrail violations even when configured with `PERMISSIVE_CONTENT_TRANSFORMATIONS`. I'd like to submit these to help improve the models, but not having a Python API I can call from the notebook means I probably won’t, and if I cannot find a workaround, I will need to switch to using MLX and a different model.

[1] https://developer.apple.com/documentation/foundationmodels/improving-the-safety-of-generative-model-output