Refactor into base tensorflow component

[getnamo]

We can probably add a tickbox or a tensorflow component subclass which would enable remote tensorflow. This would work via re-routing calls to the tensorflow component to e.g. a python socket.io server which would call the same TFPluginAPI scripts on the server and return results without changing any blueprint or python API currently used.

This would enable easy scaling for training or production if greater performance is needed without depending on it locally.

[getnamo]

Likely best to separate into a new plugin (https://github.com/getnamo/tensorflow-remote-ue4) so that it does not have UnrealEnginePython dependency. We can use that plugin as a dependency to define the base class tensorflow component with C++ calls so both share same API.

[getnamo]

base class will be in remote tensorflow plugin which this will depend on. Remote tensorflow plugin will only depend on socket.io plugin.

Variant summary:

• https://github.com/getnamo/tensorflow-ue4 depends on
  -> UnrealEnginePython
  -> SocketIOClient
• https://github.com/getnamo/tensorflowjs-ue4
  -> NodeJs
  -> SocketIOClient
• https://github.com/getnamo/tensorflow-remote-ue4
  -> SocketIOClient
• https://github.com/getnamo/tensorflow-native-ue4
  -> Tensorflow C Lib

will later have

• https://github.com/getnamo/tensorflow-ue4 depends on
  -> UnrealEnginePython
  -> TensorflowRemote
  -> SocketIOClient