PROPFIND href segments were decoded with java.net.URLDecoder, which applies the
form-encoding rule '+' -> space. A file named '6103+.pdf' was read as '6103 .pdf',
so every GET/DELETE requested '6103%20.pdf' and 404'd forever — that one file could
never sync. Replaced with decodeWebDavSegment(): decodes %XX only, leaves '+' literal.
Covered by WebDavDecodeTest (7 tests).
