Kalman Filter Optimizer

[Leo-VK]

This is what I implemented so far concerning the kalman filter with the least squares loss. Note that I did not externalize the hessian calculation to the losses.py module yet, since the we do not need to assemble the hessian directly if the shermann-morrison formula is used.
Things that still need consideration are:

• Efficient implementation for general losses
• Neat integration/consistent signature for _update_after_pred