Replace the analytical approach of comparing linestrings by an empirical one.

Fixes slow (>1 sec) behavior when ranking large routes.
The empirical approach is faster by a magnitude of about three for large routes (> 100 km).