Faster CSG operations

[BrunoLevy]

• [ ] fast_union mode, make it properly handle connected components floating inside another primitive
• [ ] use 64 bits instead of 32 for primitive inclusion bits
• [ ] tree transforms: create larger trees (what is a good balance ? we could completely "flatten" the tree)
• [ ] tree transforms: find common subexpressions that could be factored (DAG rather than tree), see how the cache works in OpenSCAD.