Optimize performance of check_clearance_copper

based on #650

Main optimisations:
It is now ensured that every patch that needs expanding will only be
expanded once. Additionally the number of unique patches actually
expanded should be reduced.

The later is achieved by a primitive implementation of set-cover. Which
is running a single pass over all patch_pairs and if no patch is
already expanded by the required clearance the patch with more
neighbours is used.

Additionally only patch_pairs that have overlapping bounding boxes will
be scheduled for parallel checking of pair intersection. This had
the upside of also calculating the bounding boxes only once.

Other optimisations:
- Pre-filter patches and calculate patch pairs in a more structured
approach.
- Trusting the bounding box check more, not using twice the clearance
only clearance + 10 (Even though I don't see the point in even doing
that, but carrotIndustries insisted...).

Bug fixes on the way:
- Copper around a via is no longer matched against copper on all other
layers
- Intersection PTHs will be reported. This is that big a deal as PTHs
should always be surrounded by copper which reported the error of e.g.
overlapping vias previously.

Author: carrotIndustries