Calculate Jacobians with forward AD from num-dual

[prehner]

Hello,

this is in reference to #24 . Jacobians are calculated using forward mode automatic differentiation provided by the num-dual crate. The user only needs to implement the residuals using the DualNum trait bound, everything else comes for free (implementation-wise). Because, with AD the residuals and Jacobians are always evaluated simultaneously, I simplified the interface for the LeastSquaresProblemAD a bit compared to the LeastSquaresProblem trait. Also, num-dual uses the Owned data storage in nalgebra, so this is also hardcoded here, but might be relaxed if that is necessary.

To validate, I adapted the line fitting example in tests. This is still a bit raw and I am happy to clean it up and document it properly if you are interested in having this feature.