fix: guard ModelNameCompleter against None active_model and stale cache

[varunsahni18]

Changes

• Guard get_active_model() return against None before calling .lower() (fixes AttributeError crash)
• Remove stale self.model_names cache from __init__
• Load model_names fresh per completion invocation in get_completions()
• Add regression test for the no-model-configured scenario

Issue

Closes #419

Testing

from unittest.mock import patch
from code_puppy.command_line.model_picker_completion import ModelNameCompleter
from prompt_toolkit.document import Document

with patch("..._load_models_config", return_value={"gpt-4": {}}), \
     patch("...get_active_model", return_value=None):
    completions = list(ModelNameCompleter().get_completions(Document("/model "), None))
    assert len(completions) > 0  # No crash!

Credit: Issue analysis and code pointers by @mpfaffenberger.