Add support for lower than 8-bits precision

This MR intend to add support for lower than 8-bits precision. It extends on the first PoC done on export_arm_cortexm and generalizes it for export_cpp:.

TODO:

• core: Add conversion support for lower than 8-bits data types;
• core: Extend/improve applyWeightInterleaving recipe; check later
• export_cpp: Adapt kernels to support lower than 8-bits data types;
• export_cpp: Add lower than 8-bits unit tests.