Implement a Standardized API to access Models

[MaxenceGui]

Issue description

To ensure a scalable and maintainable system, every model should have a standardized access method. Currently, the system needs to adapt three ways of accessing three models.

Expected Behavior

Models are accessed using the same request format each time, with a consistent header and body. The api_key and endpoint_url parameters are expected to be unique for each model.

Current Behavior

Currently, there are three different ways to call three models. Here's a list of functioning headers and body by models

Nachet-6seed:
headers = {
    "Content-Type": "application/json",
    "Authorization": ("Bearer " + endpoint_api_key),
}

body = {
    "input_data": {
        "columns": ["image"],
        "index": [0],
        "data": [image_bytes],
    }
}

Seed-detector:
headers = {
    "Content-Type": "application/json",
    "Authorization": ("Bearer " + endpoint_api_key),
    'azureml-model-deployment': 'seed-detector-1'
}

body = {
    "input_data": {
        "columns": ["image"],
        "index": [0],
        "data": [image_bytes],
    }
}

Swin:
headers = {
   "Content-Type": "application/json",
   "Authorization": ("Bearer " + api_key),
}

body = b64encode(image)

Additional Context

Issue 51 Implement Model Switching Functionality in Inference Request

[rngadam]

@amirardalan9473 can you provide an explanation and link to documentation on how the app.py/score.py is selected and inserted for the model.

"standard way" should be standardized API

also needs discussion on what functionality and data format we want for that API