MicroMix: Efficient Mixed-Precision Quantization with Microscaling Formats for Large Language Models