Implement LLGuidance from upstream

[iwr-redmond]

Support for LLGuidance, which uses constrained sampling to facilitate valid JSON output, has recently been added to llama.cpp. It's the difference between asking "pretty please" and guaranteeing valid output.

Enabling this feature during compilation requires some fiddling with Rust, and there are still some bug fixes that need applying (see pull 11644). However, this would make it very easy for end-users to obtain validly structured responses from models, and is worth considering for easy-llama, as it would eliminate the need to work with a third-party library like LM Format Enforcer.