Significant optimization: scans up to twice as fast for some classpaths.
For jars where both the root of a classpath element (jar or directory) and a subdirectory like WEB-INF/classes are both on the classpath, classpath scanning in the ancestral classpath element will stop recursing at the nested classpath element. This eliminates duplicated scanning, and can make scanning up to twice as fast for those scenarios. (#127 -- thanks to @DavyDeDurpel for reporting.)