Add min_sqnr to torchao/_models (#1502)

* Add min_sqnr to torchao/_models

Summary:
att, this allows us to run benchmarks with different min_sqnr to control the accuracy
loss of the model

Test Plan:
```
export CHECKPOINT_PATH=checkpoints
export MODEL_REPO=meta-llama/Meta-Llama-3.1-8B
python torchao/_models/llama/generate.py --checkpoint_path "${CHECKPOINT_PATH}/${MODEL_REPO}/model.pth" --compile --compile_prefill --quantization autoquant-all --output_json_path benchmark-results.json --outpu\t_json_local --min_sqnr 40

python torchao/_models/sam/eval_combo.py --coco_root_dir datasets/coco2017 --coco_slice_name val2017 --sam_checkpoint_base_path checkpoints --sam_model_type vit_h --point_sampling_cache_dir tmp/sam_coco_mask_ce\nter_cache --mask_debug_out_dir tmp/sam_eval_masks_out --batch_size 32 --num_workers 8 --use_compile max-autotune --use_half bfloat16 --device cuda --compress autoquant-all --output_json_path benchmark-results.json\ --output_json_local --min_sqnr 40

cd examples/sam2_amg_server
python server.py ../../${CHECKPOINT_PATH}/sam2 large --port 4000 --host localhost --fast --use_autoquant --benchmark --dry --output_json_path ../../benchmark-results.json --output_json_local --min_sqnr 40
```

Reviewers:

Subscribers:

Tasks:

Tags:

* change origins

* add min_sqnr to result

* add min_sqnr to result

* cleanup