#37 Add the ability to compute part of the jacobian numerically

[Helfer Thomas]

Computing the jacobian of the implicit system is the most difficult
part of implementing a behaviour. Computing the jacobian by finite
difference is interesting but significantly decreases the performances
of the behaviour and can be (very) sensitive to the choice of the
numerical perturbation.

The @NumericallyComputedJacobianBlocks keyword is used select a list
of jacobian blocks that have to be computed numerically. This is more
efficient than computing the whole jacobian numerically. Combined with
the ability to compare the jacobian to a numerical approximation, the
user now has the ability to build the jacobian incrementally, block by
block and checks at each steps that their analytical expressions are
correct.

This keyword can optionnaly be followed by a list of modelling
hypotheses. The list of jacobian blocks is given as an array.

Notes

• This keyword can be used multiple times. The newly declared jacobian
  blocks are added to the existing ones.

Example

@NumericallyComputedJacobianBlocks {dfp_ddeel,dfeel_ddeel};