feature: Native support for /v1/messages endpoints

Did you check the docs?

• I have read all the NeMo-Guardrails docs

Is your feature request related to a problem? Please describe.

While there is a langchain integration with /v1/messages, there is no external endpoint that accepts /v1/messages payloads. This means it's impossible to directly proxy a /v1/messages endpoint - you'd have to translate your request into /chat/completions first, then translate the response.

I'd like to be able to natively use /v1/messages throughout the library, namely as models specified in the model config (for self-check, etc), or to use the API as my main interface for interacting with NeMo Guardrails.

Describe the solution you'd like

• A native /v1/messages client, mirroring the existing /v1/chat/completions client, that allows for native, non-langchain communication with /v1/messages models
• A /v1/messages endpoint in the NeMo Guardrails server, for guardrailed inference proxying
• A /v1/messages/checks endpoint, for standalone guardrailing

Describe alternatives you've considered

External API translation by some other middleware component. This works, but adds setup complexity and can present issues with streaming.

Additional context

No response

Start by locating the existing /v1/chat/completions client and NeMo Guardrails server endpoint, then trace how model configuration and standalone checks use them. Define the matching behavior for /v1/messages, /v1/messages/checks, and streaming before implementing the client and proxy paths; done means native payloads work without translation and the existing checks remain available.