Abstract duplicate code in TensorFlowTensor & TensorFlowVariable

[jvmncs]

To do this properly, we'll want to move some of the code in TorchTensor from syft/frameworks/torch/tensors/interpreters/native.py into the AbstractTensor at syft/generic/tensor.py, and then call most of our code for these classes from the AbstractTensor base for TFTensor & TFVar.

[jvmncs]

KerasObject seems to fulfill similar functions as well -- the class hierarchy here should probably be reviewed