Architecture-Compiler Co-Optimization of Computing-in-Memory Edge-AI Accelerators