feature: Tool call validation for v1/chat/completions requests

Did you check the docs?

• I have read all the NeMo-Guardrails docs

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

While the tool_choice and tools params exist in the GuardrailsChatCompletionRequest schema, their behaviour is not perfectly aligned with the OpenAI Chat Completions API. Broadly speaking, there is no currently no validation of the tool_choice param in GuardrailsChatCompletionRequest. In contrast, OpenAI Chat Completion API checks the tool_choice argument in the incoming request with the list of tools. This ensures that all tool choices declared in chat completions requests are valid. In order to align tool-calling behaviour in GuardrailsChatCompletionRequest with OpenAI API Chat Completions API, initialization and run-time validation are necessary.

Describe the solution you'd like

• Optional configuration of tools on config.yaml to define a list of tools the model can access i.e adding a new field, tools in LLMRails. This is meant to mimic OpenAI's Tool Search where we load all the available tools at server start time rather than having the user send tool definitions at runtime
• Check if the requested tool_choice in a guardrails v1/chat/completions is available in the list of tools
• Enrich the existing tools and tool_choice params. with missing subfields

Describe alternatives you've considered

n/a

Additional context

No response

Start with the GuardrailsChatCompletionRequest schema and the LLMRails configuration path, then trace handling for tools and tool_choice in v1/chat/completions. Confirm how config.yaml could define server-side tools, where initialization and run-time validation belong, and which missing subfields need to be represented. Done means invalid tool choices are rejected and valid configured tools align with the OpenAI Chat Completions behavior.