Support customizing the type of quantization

[chenbohua3]

Different backends have different types of quantization, such as:

• per tensor / channel quantization
• uint / int quantization

Currently the type of quantization is hard-coded in each quantizer which is not flexible. So is there any needs to customize it by user setting? A potential way is to set through configure_list like:

configure_list = [{
            'quant_types': ['weight', 'input'],
            'quant_bits': {'weight': 8, 'input': 8},
            'op_names': ['conv1'],
            'dtype': 'uint',  # or int
            'qscheme': 'per_tensor_affine' # or 'per_tensor_symmetric', 'per_channel_affine, 'per_channel_symmetric'
        }]

The keyword dtype and qscheme and corresponding values are derived from PyTorch quantization which may be familiar to PyTorch users