Add general bregman divergence dre

[thomvolker]

Use optim function to let users specify their own loss function

[thomvolker]

This can also be done using torch or a similar library.