Extend exp_overapproximation API

[schillic]

Since #118 and #144 we have two alternative algorithms to compute the matrix exponential. We should extend the exp_overapproximation to choose the algorithm. Note that there is an additional input parameter. Maybe the API should instead create a wrapper object holding the parameters.

We should also make the Horner scheme available in the scaling-and-squaring algorithm.

@mforets also wants to rename the function scale_and_square.