Add OutputVector class and a more complex example to Tensorflow

[Neiko2002]

I have created a smaller version of the ExampleTrainer script and was working on a complete training example. In order to finish it some additional features were needed. e.g. the AddN constructor needed an InputList which itself requires a vector of outputs. The vector does not exists right now, as far as I know.

public AddN(@Const @ByRef Scope scope, @ByVal InputList inputs) 
...
public InputList(@StdVector Output out)

The sames goes for the AddSymbolicGradients method

public static native @ByVal Status AddSymbolicGradients(@Const @ByRef Scope scope,
                            @StdVector Output outputs,
                            @StdVector Output inputs,
                            @StdVector Output grad_outputs)

This PR creates a OutputVector class and adds three examples: a simpler version of the ExampleTrainer script, a InputListExample and the more sophisticated examples requested in issue #409.

[saudet]

Looks good, thanks! Since the samples are not run as tests though, could we put them under the samples subdirectory in their own little project with a pom.xml file to make them more accessible? (FYI, I'd eventually like to run samples automatically as part of tests, but don't worry about that for now.)

[Neiko2002]

There is now a sub directory in tensorflow called "samples" containing its own maven project. All code samples are there.