We were using md5_hex to generate a unique file id from a full path filename. Changed this to use the file device and inode id instead. Should still provide unique file ids and the file names are shorter too. Fixes #29