Pytorch inference fixes

[beniz]

Small fixes and improvements to torch model inference:

• return all classes with "best":-1
• cleaner error message when model location is wrong
• allowing std as a vector of float
• introducing scale as an option to scale the image values

The torch image classification model service creation and prediction become:

• service creation

curl -X PUT "http://localhost:8080/services/torch_resnet" -d '{
    "description": "image classification service",
    "mllib": "torch",
    "model": {
        "repository": "/path/to/resnet50_torch/"
    },                   
    "parameters": {
        "input": {                                                            
            "connector": "image"
        },"mllib":{}
    },              
    "type": "supervised"
}     
'

• prediction

curl -X POST "http://localhost:8080/predict" -d '{
    "service": "torch_resnet",
    "parameters": {
        "input": {
            "width":224,
            "height":224,
            "rgb":true,
            "scale":0.004,"mean":[0.485,0.456,0.406],"std":[0.229,0.224,0.225]
        },"mllib":{"gpu":true,"gpuid":0},
        "output": {
            "best":3
        }
    },
    "data":["/path/to/cat.jpg"]
}
'

This is using the mean and std values from https://pytorch.org/docs/stable/torchvision/models.html