chore: bump mlx-swift-lm to pick up the Qwen3VL unit tests (#148)

Points at 3f6f13e (SharpAI/mlx-swift-lm#50), which adds the second set of unit
tests for anything under MLXVLM — Qwen3VL.swift, 1770 lines, the largest
zero-coverage file in the model layer after #49/#147 covered Gemma4.

No SwiftLM-side code changes; verified the umbrella builds clean against the
bumped pointer.

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

Author: solderzzc