Add test suite for approximation code, fix edge cases

[hannobraun]

The approximation code queries the geometry for approximate points, then processes and aggregates these points, depending on the type of edges.

I've cleaned this up recently, but I suspect there's at least one unaddressed edge case in there. We should have a comprehensive test suite for this code, possibly move it to a dedicated module, and make sure we cover edge cases as best as possible.

---

Here's the unaddressed edge case that I suspect needs addressing.

Point de-duplication depends on consistent edge direction

When approximating an edge, the approximation will include the two points that bound the edge. When approximating a whole cycle of connected edges, this will lead to duplicated points that need to be filtered out.

Currently, the points are in a Vec and Vec::dedup is called to filter out duplicates. However, this won't work if the direction of edges is inconsistent. For example, if you have the straight edges a -> b, b -> c, then the duplicate b will be correctly filtered. If the edges are a -> b, c -> b, on the other hand, then one of the bs must still be filtered, but that won't happen.

[hannobraun]

Turns out the edge case I mentioned here was worse than I thought: The problem would even show up when edge direction was perfectly consistent, as the first vertex and the last vertex weren't recognized as duplicates (even though, the edges forming a cycle, those were identical).

It's been fixed in #158.