[Feature]: Add a guided CLI wizard for goal-driven model recommendations

### Problem or motivation

LLM Fit already does a great job of evaluating models against a user's hardware and provides use-case-specific recommendations. However, using these capabilities still requires the user to understand concepts such as model categories, use cases, hardware constraints, and the available CLI options.

For a non-technical or first-time user, the starting point is usually not:

_"Which model should I run for the coding use case?"_

Instead, it is more likely to be:

_"I want a local AI model that can help me summarize PDFs and answer questions about them. What should I use on my computer?"_

There is currently a gap between what the user wants to accomplish and how LLM Fit expects that intent to be expressed through its existing interfaces.

A guided CLI experience could make the existing recommendation engine more accessible without changing or duplicating the underlying model-fitting logic.

### Proposed solution

Add an optional interactive CLI wizard, for example:

**llmfit wizard**

The wizard could ask a small number of simple questions about the user's intended workload and priorities, such as:

What do you want to use a local AI model for?
How important is response speed versus model quality?
Do you need a large context window?
Do you need vision/multimodal capabilities?

The wizard would then translate these answers into the existing LLM Fit recommendation/use-case system and return a ranked list of suitable models for the user's detected hardware.

For example:

**$ llmfit wizard**

What do you want to use a local AI model for?

1. General assistant
2. Coding
3. Documents / PDFs
4. Reasoning
5. Writing
6. Images / multimodal

> 3

What matters most to you?

1. Best quality
2. Fast responses
3. Balanced

> 3

Analyzing your hardware...

Recommended models:

1. Model X — Q4_K_M
Best overall fit for your hardware and document workflow

2. Model Y — Q5_K_M
Higher quality, but slower

3. Model Z — Q4_K_M
Faster alternative

The important part is that this would be a UX layer over the existing recommendation capabilities, rather than a replacement for the current recommendation engine.

It could also remain completely optional, so existing users can continue using the current CLI/TUI workflows.

### Alternatives considered

A few alternative approaches could also be considered:

Extend the existing recommend command with an interactive mode rather than introducing a separate wizard command.
Add a **--interactive** flag, for example:
llmfit recommend **--interactive**
Expand the existing TUI to provide a guided recommendation flow.
Keep the CLI unchanged and expose the same functionality through the existing web interface.

I would personally favor an interactive CLI flow as the initial implementation because it keeps the feature lightweight, terminal-friendly, and relatively isolated from the existing recommendation engine.

### Feature area

CLI (new subcommand or flag)

### Would you be willing to contribute this?

Yes, I'd like to submit a PR

### Additional context

_No response_

Start by reading the existing CLI entry points, especially the current recommend command and the recommendation/use-case system it calls. Decide whether the UX should be a new llmfit wizard subcommand or an interactive flag, since the issue leaves that open. Done means the interactive flow asks the proposed goal/priority questions, maps answers to existing recommendations, and returns ranked models without replacing the current engine.

Rédigé par le modèle d'indexation à partir du texte de l'issue.