Refactor Fit objects

[tBuLi]

This PR represents a mayor overhaul of the internal structure of Fit objects in symfit.

Fit is now responsible for feeding an Objective function to a Minimizer, it is no longer the combination of these two.

This means two new types of object have been introduced; Minimizers and Objectives. This change should make it easier for people to introduce a new Minimizer while being completely agnostic about which objective function to use.

To do (not a complete list, work in progress):

• Documentation overhaul
• Developers instructions in the docs
• Tests
• Make keyword arguments consistent across minimizers (e.g. #106 )
• Covariance matrix for all types of models including Log-likelihood (#138)
• Wrap all minimizers in scipy.optimize.minimize with the same new API.
• ...

[pckroon]

We need to address one more thing in this fell swoop, and that is to decide how the arguments are called that fit.execute (and thereby Minimizer.execute) is to accept, and what it should return. In the process we should standardise some names.

[tBuLi]

Note: The error that now remains in this branch is one that will be solved my merging with @pckroon's finite differences branch, because then stdev's for ODEModels can be calculated.

[pckroon]

Looking good! Note that you should probably fix #56 before tagging and releasing.