mlx-native 0.13.0

What's Changed

• fix(dense): execute short bf16 reductions natively by @robertelee78 in #37
• feat: preserve native GGUF execution across model lifetimes by @robertelee78 in #38
• test(q5_0): gate optional tensor artifact by @robertelee78 in #39

Full Changelog: v0.12.2...v0.13.0