Tweaks for running on Linux

[donn]

This commit contains a number of tweaks to ensure exporting quantization and testing inference both work on Linux-based setups.

• exportquant.py: made np.uint32 data type explicit -- on some configurations, the type int appears to be bridged by numpy into np.int64, so when the weights are packed, they're packed into 8 bytes instead of 4, and when the view is converted to np.uint32 every other entry is just 0x00000000.
• BitNetMCU_MNIST_dll.c: added forward declaration for BitMnistInference and a guard for __declspec (not supported by gcc)
• requirements.txt: Added tensorboard, matplotlib
• Created Makefile for compilation on Linux

[cpldcpu]

Awesome! Thank you a lot for fixing this for linux. I should look into extending the makefile to work with vs as well.

Merged!