Provable Post-Training Quantization: Theoretical Analysis of OPTQ and Qronos