A Computing-in-Memory Engine Supporting One-Shot Floating-Point NN Inference and On-Device Fine-Tuning for Edge AI