[Feature]: Decouple LLM setup from onboarding, force in chat view when model not detected

[fiorittoev]

Problem / motivation

Users should not be able to chat without first connecting a model

Proposed solution

Create a reusable second onboarding stage for setting up local llm. This should be used on initial setup after onboarding and whenever there is no connection in chat view

Scope & constraints

No response

Alternatives considered

No response

Acceptance criteria

Seperate Onboarding and LLMOnboarding into two components
Call when LLMs not detected

Test plan

Test by registering llm then disconnecting, etc.