Support prompt caching for Claude AI chat models

[KallynGowdy]

Claude has added the ability to cache input prompts to increase performance and reduce costs. The caveat is that writing to the cache costs more than a regular input token, but the benefit is that a prompt that hits the cache will save 90% over what it would cost without the cache. CasualOS should support specifying if a message can be cached, and the AnthropicAIChatInterface should pass along the correct values to the Anthropic API.