Rename some transforms using appropriate wide adopted naming conventions

🚀 Feature

Rename confusing transformation names such as transforms.Normalize() to transforms.Standardize() and decouple combined operations transforms.ToTensor() into transforms.Normalize()/transforms.Rescale() and transforms.ToTensor() for clarity.

Motivation

I'm always frustrated when trying to communicate the above operations (i.e. transforms.Normalize(), transforms.ToTensor()) to statisticians, the reason for that is that statisticians use the following conventional naming standardize = rescale data to zero mean and unit variance i.e., what transforms.Normalize() does, and, normalize = rescale data value into the range[0, 1], i.e., what tranforms.ToTensor() does, which is basically 2 operations, transforming the PIL image into a tensor and rescaling it.

Pitch

Is it possible to rename according to the world wide convention that statisticians use in order to avoid confusion in the future and have better clarity of what each operation is meant to be doing?

1. Rename transforms.Normalize() into transforms.Standardize()
2. Break transforms.ToTensor into two separate operations transforms.Normalize() which basically rescales the values into the range[0, 1], and transforms.ToTensor() which basically transforms a PIL image into a torch tensor.

cc @vfdev-5

Start by reviewing the existing transforms.Normalize() and transforms.ToTensor() APIs and their documented behavior. Compare those behaviors with the proposed Standardize(), Normalize(), Rescale(), and ToTensor() naming. Done would require an agreed migration plan and corresponding updates across the public transform API, documentation, and tests.