[Performance] ONNX runtime takes too much CPU and RAM and slows the entire PC

@yihonglyu is already working on this.

Since Apr 9, 2024.

### Describe the issue

I've converted a paddleOCR model into an onnx model , the model is around 9Mb and I run it on the CPU , but whenever I activate the onnx runtime to run the code it takes all the remaining CPU resources of the PC and a large chunk of the Ram.

### To reproduce

I'm pretty sure it has to do with the input shape logic , I've tried both static and dynamic input convergion . I've tried the onnx exporting tools Paddle2onnx and PaddleOCRModelConvert

this is the script i'm using to test the runtime.
```
ort_session = rt.InferenceSession('model.onnx')
so = rt.SessionOptions()

print(ort_session.get_inputs()[0].shape)
print(ort_session.get_outputs())

def preprocess_image(image_path):
image = Image.open(image_path).convert('RGB')
# image = image.resize((1000, 1000))
image_array = np.array(image)
image_array = image_array / 255.0
image_array = np.expand_dims(image_array, axis=0)
image_array = np.transpose(image_array, (1, 3, 0, 2))
return image_array

def get_memory_usage():
process = psutil.Process()
return process.memory_info().rss / (1024 ** 2) # Memory usage in MB

# List of image paths
image_paths = ['./ppocr_img/imgs_en/img_12.jpg']*50
memory_usage_list = [get_memory_usage()]
img_path = image_paths[0]
#for img_path in image_paths:

memory_before = get_memory_usage()

input_data = preprocess_image(img_path)
print("Input data shape:", input_data.shape)
ort_outputs = ort_session.run(None, {ort_session.get_inputs()[0].name: input_data.astype(np.float32)})[0]
memory_after = get_memory_usage()
memory_usage_list.append(memory_after)
```

### Urgency

Not urgent since it's a solo project but I spent a lot of time on this and just want to get over with it.

### Platform

Windows

### OS Version

11

### ONNX Runtime Installation

Released Package

### ONNX Runtime Version or Commit ID

1.9.0

### ONNX Runtime API

Python

### Architecture

X64

### Execution Provider

Default CPU

### Execution Provider Library Version

_No response_

### Model File
https://drive.google.com/file/d/1tCa3qzHzHEGq_KqzpAdCt5VLH-4Ym58k/view?usp=sharing

### Is this a quantized model?

Yes