Issue 33 templateize delaunay

[rgoulter]

Some notes:

• The computational geometry methods I'm using make some numbers very big very quickly.
  • e.g. the signDet method used by the "point in circle" method of Delaunay PointSet.
• So, int isn't viable as an imprecise alternative to our precise LongInt.
• double ... passes the unit tests. But it seems like there are assumptions which break.
  • e.g. apparently possible for more than one leaf node to contain the same edge. (?).

[rgoulter]

Thoughts/concerns that the code can fail-an-assert?

I guess I'd hoped that using double instead of LongInt would make it much faster.

It would be 'wrong' to "break" the GUI like this. -- It would be 'acceptable' if the quick / broken one were a choice.

In any case, I think this PR is loosely an improvement. And that the project has other issues besides speed. -- Probably offering which 'algorithm' to use would be a good idea. (And probably implementing Fortune's would be a better idea. -- Or making the Delaunay one ... visual / informative).