adding quant_format, mantissa, and exponent options to evaluate script

[fabianandresgrob]

Adds the following options for the ptq_evaluate.py script:

--quant_format {int,float}
                      Quantization format to use for weights and activations
                      (default: int)
--layerwise-first-last-mantissa-bit-width LAYERWISE_FIRST_LAST_MANTISSA_BIT_WIDTH
                      TODO
--layerwise-first-last-exponent-bit-width LAYERWISE_FIRST_LAST_EXPONENT_BIT_WIDTH
                      TODO
--weight-mantissa-bit-width WEIGHT_MANTISSA_BIT_WIDTH
                      TODO
--weight-exponent-bit-width WEIGHT_EXPONENT_BIT_WIDTH
                      TODO
--act-mantissa-bit-width ACT_MANTISSA_BIT_WIDTH
                      TODO
--act-exponent-bit-width ACT_EXPONENT_BIT_WIDTH
                      TODO

Also changed scale_factor_type argument so ptq_common.py handles weight_scale_type correctly. Still missing good explanations for the added options.