feat: add macOS Metal GPU detection for local embeddings

[lexgenius]

Summary

Currently the plugin only probes for CUDA (NVIDIA) and Vulkan (AMD/Intel) on Linux. On macOS it falls back to CPU-only, even though node-llama-cpp supports Metal natively.

This PR adds Apple Metal GPU detection to _detect_gpu_backend() in both execute.py and helpers/qmd_client.py.

Changes

• Added import sys to helpers/qmd_client.py
• Added Metal detection block to _detect_gpu_backend() in both files:
  • Checks sys.platform == darwin
  • Verifies Metal framework exists at /System/Library/Frameworks/Metal.framework
  • Returns metal to enable GPU-accelerated embeddings on Apple Silicon and Intel Macs

Detection order (after this PR)

1. CUDA — Linux/NVIDIA
2. Vulkan — Linux AMD/Intel
3. Metal — macOS Apple Silicon + Intel Macs (new)
4. CPU fallback

Testing

• Verified single Metal block insertion in each file (no duplicates)
• Detection runs only on darwin platforms, no impact on Linux behavior