CORTEX_M: Quantization configuration

The current quantizer only quantizes to int8 and does not allow any customization. The composable quantizer pattern used is already quite easy to modify and use, so one suggestion is to add a thin layer of configuration when initalizing the cortex_m_quantizer which allows the user to easily select common quantization strategies, and leave it to advanced users to directly build their own composable quantizer which can do whatever they want to.

Deciding the exact API is part of the ticket, but it will likely include a quantization config similar to the arm backend, so that when you create the quantizer it is configured like this:

int8_quantizer= CortexMQuantizer(int8_quant_config)
int16_quantizer=CortexMQuantizer(int16_quant_config)

And all ops would use either dtype where supported.

Ideally this would be coordinated with an upgrade to the arm quantizer to unify these two to be more similar in general.