Python3.9 support

This supports Python3.9 + Numpy 1.19

Sync'd to latest TFLM

Sync'd to latest TFLM

Use built-in memset and memcpy in CMSIS

This uses:
#define memset __builtin_memset
#define memcpy __builtin_memcpy

on the CMSIS-NN library to improve latency

Initial release

1.0.0

more updates for pip package