Optimize Parallelism

[mistafunk]

for many initial shapes and complex rules it is counter-productive to use both parallelism on generate level and within the shape tree generation -> only use multiple threads to generate the shape tree if there is only one input shape