v0.1 — Phase 1 baseline

Float32 runtime, ONNX→.nk conversion, 59 embedded + 49 ONNX parity tests, timm backbone parity, CMSIS wired with reference fallback.

Documentation

• Getting Started
• Testing
• Kernels

Smoke commands

make test          # 59 C++ embedded + 49 ONNX parity
make test-python   # ONNX parity + AOT compile tests

What's in Phase 1

• .nk load (file or embedded) with MLP/CNN + fused blocks (ResNet, MobileNet, ConvNeXt)
• Depthwise conv, per-axis padding, residual adds
• Optional CMSIS-NN / CMSIS-DSP backends with reference fallback
• Python packager: ONNX → .nk, timm backbone pack, AOT embed