Add lossless_decode subsection alongside memory_mode (PR #7)

Add Hardware envelope section: memory_mode=relay opens 8-16 GB Apple Silicon Macs (PR #6)

note: all 6 Lance task families validated as of 2026-05-21

Initial upload: Lance_3B image specialist, MLX bf16 (LLM + ViT + Wan2.2 VAE bundled)

initial commit