Running local LLMs on 16GB of unified memory

Placeholder post — swap in your real notes on Ollama, LM Studio, and Q4_K_M quantization tradeoffs.