[added !932] Feature/generator speed up

I found a way that makes loading modules much faster.

The main point is to not call cmake every time. Especially not for modules, where nothing has to be compiled. (There was up to half a second overhead for each module loaded.)

I get a speedup for a simple Laplace problem from 17.6s to 1.5s total runtime.

Added also a .noconfigure file after the first configuration run, which alone reduces around 14s of the startup time.

Remark: This .noconfigure has to be removed for reconfiguration of dune-py - we discussed this in the context of removing the CMakeCache.txt. This is open and must be performed somehow by pip.