quark.onnx.tools.convert_fp32_to_bfp16

Convert tensor float type in the ONNX ModelProto input to tensor bfp16.

Use the convert_fp32_to_bfp16.py to convert a float32 model to a bfp16 model:

` python convert_fp32_to_bfp16.py --input $FLOAT_32_ONNX_MODEL_PATH --output $BFP_16_ONNX_MODEL_PATH `