Add paddle quantize model support for ORT, TRT and MKLDNN deploy backend

[yeliang2258]

PR types

New features

PR changes

Others

Describe

Add paddle quantize model support for ORT, TRT and MKLDNN deploy backend

Running a quantized model is the same as running a non-quantized model and calling the interface, without any special settings

ORT

option = fd.RuntimeOption()
option.use_ort_backend()
option.set_model_path("yolov5s_new_format/model.pdmodel", "yolov5s_new_format/model.pdiparams")

runtime = fd.Runtime(option)
data = np.random.rand(1, 3, 640, 640).astype("float32")
result = runtime.infer({"x2paddle_images":data})

MKLDNN

option = fd.RuntimeOption()
option.use_paddle_backend()
option.set_model_path("yolov5s_new_format/model.pdmodel", "yolov5s_new_format/model.pdiparams")

runtime = fd.Runtime(option)
data = np.random.rand(1, 3, 640, 640).astype("float32")
result = runtime.infer({"x2paddle_images":data})

ONNX-TRT

option = fd.RuntimeOption()
option.use_trt_backend()
option.use_gpu()
option.set_model_path("yolov5s_new_format/model.pdmodel", "yolov5s_new_format/model.pdiparams")

runtime = fd.Runtime(option)
data = np.random.rand(1, 3, 640, 640).astype("float32")
result = runtime.infer({"x2paddle_images":data})

[yeliang2258]

After this PR：PaddlePaddle/Paddle2ONNX#906

[jiangjiajun]

添加单测，参照 https://github.com/PaddlePaddle/FastDeploy/blob/develop/tests/eval_example/test_ppmatting.py