Static quantized resnet is slower than the raw one

@yufenglee is already working on this.

Since Sep 29, 2020.

**Describe the bug**
Static quantized resnet is slower than the raw one.

Compared the static quantized Resnet model and the raw one from the E2E example code.
here is some benchmark from my computer

resnet50_v1.onnx / 3.03125
resnet50_v1.quant.onnx / 6.484375

here are the code piece
```

def perf_test(onnx_path, num=10):
sess = onnxruntime.InferenceSession(onnx_path)
name = sess.get_inputs()[0].name
data = np.random.random((1, 224, 224, 3)).astype(np.float32)
ort_in = {name: data}
# warm up
sess.run(None, ort_in)

latency = 0
for i in range(num):
start_32 = time.process_time()
ort_out = sess.run(None, ort_in)
latency = latency + time.process_time() - start_32
return latency

def latency_benchmark(num=10):
models = [
"resnet50_v1.onnx",
"resnet50_v1.quant.onnx",
]
latency_dict = {model: perf_test(model, num) for model in models}

for m, l in latency_dict.items():
print(f"{m} / {l}")

```

**Urgency**
Quantization performance

**System information**
Windows server
onnx==1.7.0
onnxruntime==1.4.0
Python version: 3.6

**To Reproduce**
See bug description