Merge options for lazy matrix exponential in discretization

[schillic]

We have the options :lazy_expm and :lazy_expm_discretize. The latter controls whether we use a lazy matrix exponential during discretization. The former controls whether we convert the matrix to a concrete matrix after discretization (see here).
Having the second option seems unreasonable because we could already use the concrete matrix exponential in the discretization. (Careful when removing that function, as it calls another function in the end. This other function should then be called directly from the call site.)

Furthermore, we have the option :pade_expm. We could merge these settings by changing the value to a string. The available choices would then be "none" (default), "exp" (in the previous Julia versions the function was called expm), and "pade".

[mforets]

Having the second option seems unreasonable because we could already use the concrete matrix exponential in the discretization.

If i recall correctly, that option was a workaround for the fact that we use the same "exp_method" in the calculation of two different problems: ϕ and Phi2Aabs. But the latter is more costly than the former, since it is 3 times bigger. So the idea was to do lazy computations for the discretization and then transform to concrete ϕ in the iteration.

The code in that part (#503) is now:

    # compute matrix ϕ = exp(Aδ)
    ϕ = exp_Aδ(A, δ, exp_method=exp_method)

    # compute the transformation matrix to bloat the initial states
    Phi2Aabs = Φ₂(abs.(A), δ, exp_method=exp_method)

And it still doesn't let to choose between different exponentiation methods in ϕ and in Phi2Aabs.

I think that we can continue like this, and eventually add this option if needed.

EDIT: i've opened #506 for the continuation