further de-duplication

[smnorris]

The intent of the processing is to over-estimate roads but it might be possible to add more relatively simple queries to remove obvious duplication
Eg:

• after insert, compare lengths/counts of inserted features after being split up by the buffers and delete features that are obvious duplications (short relative to source feature, endpoints of features are still close (<10m? <20m? <50m?) to higher priority roads, high count of features inserted compared to the original feature)
• compare each lower priority line to its nearest neighbour, either splitting the lines up into chunks and comparing the distance from each lower priority vertex to higher priority vertex and scoring, or breaking higher priority lines at endpoints of low priority dangles and comparing nearest neighbours with Hausdorf/Frechet distances