Update the opengl interop code to be significantly faster with cuda.

Now that we hold the state information about the transferring of an array, we can do a far more efficient transferring.