[Embedded] [Cortex-M] Improve performance of quantized MV2 or 3 performance using Arm CMSIS-NN

[digantdesai]

• Building up the existing backend/cortex-m by integrating CMSIS-NN
• Use Quantized MV2/3 as a pilot models
• Use FVP (if not a real board - preferred with M55 or M85) to showcase performance improvements over portable (reference) operators running this
• [stretch] compare against TFLM
• [stretch] compare against ET Ethos

[digantdesai]

• Got quantized Add working.