onnx bench rework

[Human9000-bit]

• Add onnx model criterion benches
• ONNX BENCH REWORK !!!!111!!

there were 2 onnx benches:

• new one, uses criterion and various
• old one, used for onnx models profiling

Those 2 guys had some shared logic, which has been moved into benchmarks/onnx-models/src/lib.rs

updated CI as well

New commands to run benches:

• Criterion:

cargo bench --manifest-path benchmarks/onnx-models/Cargo.toml --bench onnx_models

• Detailed profiling:

cargo run --release --manifest-path benchmarks/onnx-models/Cargo.toml --example bench_yolo

[Human9000-bit]

/bench

[github-actions]

ONNX CPU benchmark

GitHub-hosted runner signal, advisory only. Base and head are compared within the same architecture job.

arch	case	base p50	head p50	delta

Head op/kernel summary

arch	case	ops	dispatched kernels

Missing paired result(s): aarch64/gemm_relu_256x512x512, aarch64/mobilenet_v3_small_224, aarch64/resnet18_224, aarch64/small_conv_relu_64, aarch64/small_dw_pw_64, aarch64/small_gemm_relu_1x128, aarch64/small_residual_conv_64, aarch64/winograd_3x3_20x20_c256, aarch64/winograd_3x3_80x80_c64, aarch64/yolo11n_bus_640, aarch64/yolov8n_bus_640, x86_64/gemm_relu_256x512x512, x86_64/mobilenet_v3_small_224, x86_64/resnet18_224, x86_64/small_conv_relu_64, x86_64/small_dw_pw_64, x86_64/small_gemm_relu_1x128, x86_64/small_residual_conv_64, x86_64/winograd_3x3_20x20_c256, x86_64/winograd_3x3_80x80_c64, x86_64/yolo11n_bus_640, x86_64/yolov8n_bus_640

Raw JSON artifacts include per-run min/p50/avg/p95/p99, optimized node counts, graph-cost summaries, yscv CPU dispatch report, and per-node runner profile summaries.

[Human9000-bit]

ig we need to merge this for bench to work