Refactoring of the vqs class.

[RehMoritz]

The changes are mostly related to symmetrizations of the wave function.
It is now required to wrap the ansatz in the sym_wrapper utility (this of course still allows the absence of any symmetries) before passing it to the vqs class.

Also ansatz functions consisting of two networks (for amplitude and phase) are now treated differently and need to be wrapped in the TwoNets class before wrapping the returned object using the SymNet class contained in sym_wrapper.
This makes it such that the NQS class will now always receive a single network that it can evaluate, leading to a clearer structure.

In this process changes to the init_net function contained in util, aswell as numerous other changes including the tests, sampler and some of the network definitions became necessary.

All tests are passing.

[markusschmitt]

Forcing everyone to manually wrap their models in a SymNet (and potentially a TwoNets) irrespecitve of whehter symmetries are used is too cumbersome. I wrote a more flexible interface to pass nets to the NQS class.