Qwen-ssm-repair fixes Qwen 3.5 weight drift

The Qwen 3.5 weight drift illustrates how architectural complexity in hybrid SSM-Transformer models can lead to subtle failure modes that bypass standard benchmarks like NIAH. Surgical patching using statistical anomaly detection and mmap in-place patching provides a fast, cost-effective alternative to expensive fine-tuning, especially for local LLM users facing quantization-induced variance in sensitive SSM layers.