v0.0.1: Use addmm in linear operation (#27)

Initial release

This first release integrates the following fixes since the last uploaded benchmarks:

• 🐛 CUDA benchmarks are now fixed
• ⚡️ x @ y + z as linear operation in MLX is changed by mx.addmm
• the overall benchmark runs faster by spawning less processes