Replaced the simple query-based token detection with the one written by Mikhail Kotelnikov which uses a token tree of all named entities.

This is faster and also handles tokens which contain strings.
But on the other hand it does not handle substring matching yet and uses an in-memory tree of all named entities.