Consider adding an EncryptedReasoningBlock

[stephentoub]

We added TextReasoningContent, which is already supported by a variety of providers. Some AI services, however, also support encrypted reasonining content, where the reasoning information isn't sent back in plain text but rather in an encrypted form. It's provided not for the consumer but to be roundtripped back to the service, where it will be able to consume it. We should consider exposing an EncryptedReasoningBlock to abstract over such data.

[jeffhandley]

Anthropic and OpenAI both have this concept as of now.