fix: prevent AttributeError crash when Llamafile model download fails

[octo-patch]

Problem

When using interpreter --local and selecting the Llamafile provider, if download_model() fails (e.g. insufficient disk space, network error, or the user's machine can't fit any of the listed models), it returns None. The code then unconditionally executes:

model_name = model_path.split("/")[-1]

This crashes with AttributeError: 'NoneType' object has no attribute 'split' instead of displaying a clear, actionable error.

The bug affects two code paths in local_setup.py:

1. No models exist — model_path = download_model(...) in the if not models: branch returns None
2. Download new model — user selects "↓ Download new model" in the else branch; download_model() returns None; the if model_path: guard correctly skips the process launch, but execution still falls through to the crashing line

Solution

Add an early-return guard immediately before the Llamafile configuration block. When model_path is None, print a clear message and return the interpreter unchanged — instead of crashing.

if model_path is None:
    print(
        "\nNo model was downloaded. Cannot start Llamafile. "
        "Please try again or choose a different provider.\n"
    )
    return interpreter

Testing

• Normal Llamafile flow (existing model or successful download) is unchanged — the guard only activates when model_path is None.
• Simulated a failed download by having download_model() return None; previously crashed with AttributeError, now exits cleanly with a user-friendly message.