[Backend] Restrict model selection per feature for Self-Hosted LLM models

Currently we allow any registered model to be selected for either Code Generations and Code Completions. But some models can’t do both functions, e.g.: Mistral. So even though the user can select this model through the UI, it will have no effect. In this case it is an even bigger problem since the self-hosted user expects its data to be sent to its own local model, not our vendor AI provider.

UX/UI issue: #463801 (closed)

Implementation proposal

Use a yml file to store metadata about the AI features, and use these data to validate the creation of a new In the Ai::FeatureSetting. See this draft to have a full overview of the backend implemention.