Support selecting the code generation model in the UI?

## Problem
The UI has no way to choose the code generation model, and backend model selection is still tied to provider/key combinations. This can break deployments when the connected OpenAI-compatible endpoint only supports a subset of model names.

## Proposal
Expose a model dropdown in Settings and let the selected model flow through generation requests, with fallback behavior when a provider is unavailable.

I already validated a minimal approach locally:
- frontend settings UI exposes model selection
- frontend defaults to an OpenAI model when only OpenAI config is present
- backend model lists are aligned to models supported by the connected endpoint

## Question
Would you be open to supporting model selection in the UI?
If yes, I can turn my local changes into a PR and adjust them to your preferred direction.