Make type detection smarter

[tusbar]

Today types are selected if – at least – one of the identified file types matches.

We should improve that and add some sort of type detection that ignores stuff like application/octet-stream and gives priority to text/html.

For example, if a foo.kmz file returns a list of types that includes text/html (with a 200 status code) we should probably not consider it as a kmz archive.