feat: complete 4-method benchmark with honest memory reporting

feat: complete honest 4-method benchmark both models

feat: Mistral results for long context benchmark

feat: Llama results for long context benchmark

impl: Benchmark results of Llama

impl: Benchmark results of Mistral

feat: auto-save integration results to JSON

feat: sensitivity_map + bit allocation for Llama

Updated bit allocation for mistral

feat: baseline + calibration script