Allow for several systems in a single Python instance

[RudolfWeeber]

Goal

Several instances of espressomd.system should be createable in a single Python script.

Benefits

• The ML group needs this for optimizatnio tasks
• Stuff like Gibbs ensemble or replica exchange could be written down with standard Python theading and without a contorl script talking to two or more other scripts
• The test suite would run much faster, since a lot of the runtime is consumed by the numpy/scipy modules being loaded over and over again
• Getting rid of the globals in the core, which prevent this at present, is a goal in it self

Steps

• Introduce a system class in the core mirroring espressomd.System

  Note: This can initially be made globally accessible to ease the transition

• Bundle the remaining "clusters of globals" into strucutes. These are mainly a bunch of globals relating to propagation and equaiton of motion (thermostat)

• Get rid of mpi callbacks, by moving the functionality to the script interface.

  The big ones are:

  • Setting/getting particle properties (this is in script interface, but uses the mpi callbacks)
  • The reaction methods, which make use of mpi callbacks. They have to be either fully parallelized or moved to Python