TODO
----

Start/Endpoints of queries can only start on nodes only. Instead of selecting the nearest node the nearest edge should be selected.

The server part is mostly adapted from the boost http examples. It should be replaced with a self-written front-end

The KD-Tree uses a lot of RAM. An own implementation might be better.

Use Boost to compress all in-/output streams to the file system

Rework extractor to be fully externalized