Always build features so trajectories can be saved before being used

[khurram-ghani]

Always build feature-functions using a dummy shape. This is important for tf.saved_model.save to work, in cases where the user hasn't called the trajectory yet. By calling build in __init__ we ensure all the variables always exist.

[khurram-ghani]

I presume the test for this will be in @uri-granta 's PR so no need to introduce a test here?

Yes, that is correct.