feat(llm): add QMD_FORCE_CPU env var for older GPUs

[JasonOA888]

Problem

On older NVIDIA architectures (e.g., Pascal), qmd query fails with:

[node-llama-cpp] CUDA error: the function requires an architectural feature absent from the device

Workaround (Before)

Users had to run:

CUDA_VISIBLE_DEVICES="" qmd query "test"

Solution

Add QMD_FORCE_CPU environment variable:

export QMD_FORCE_CPU=1
qmd query "test"

# Or one-time:
QMD_FORCE_CPU=1 qmd query "test"

This disables GPU acceleration and forces CPU-only mode by passing gpu: false to getLlama().

Fixes #299