Improve the workflow with the `Twiss` object

[felix-andreas]

Currently an instance of the Twiss class is linked a given Lattice object. If the Lattice object changes so will the Twiss object.

This can be counter-intuitive if one wants to save some reference values before optimizing a given lattice and can lead to mistakes:

twiss = ap.Twiss(lattice)
ref_beta_x = twiss.beta_x

# do the opitmization
beta_x = twiss.beta_x

Now ref_beta_x is bound to the same underlying numpy.ndarray as beta_x. This might be error prone.

One solution to that would be to always allocate a new numpy.ndarray when calculating th twiss parameter instead of reusing the old one. Test how large the overhead would be!