Redo parallelization levels

Take into account:

1. New level between 0 and 1: "per model evaluation" for single-node setups with multiple cores
2. Level 2 could be removed, since:
   • sampling is now parallelised separately on level 1 in topologicalFilteringIteration>prepareViablePointsForProjectionSet, and
   • topologicalFiltering>topologicalFilteringRecursiveStepParallel does duplicate work when large sets of viable projections are found as in the yTor rexample (many redundant projections are processed).
3. Parallelise MC sampling from multiple points; preferably over "per model evaluation" level.
4. If done, where to put HYPERSPACE parallelisation? Preferably under MC sampling parallelisation level, whichever it will be.