Run proper trajectory in autodiff init function in torch to track gradients

[chaithyagr]

Currently we expect the trajectory is of right max value, as shown in the example_learn_samples_multicoil.py.
We want to make it more user-friendly by adding proper_trajectory in autodiff class itself and running it before piping the trajectory down to the op.