Commit provisional fix for 64 bit math overflow in libvorbisfile

seeking bisection computation

This is the equivalent of the r15921 fix in reference, but doesn't
require a double cast (for obvious reasons).  The technique is
different, the intent is the same (avoid a 64x64= >64 bit overflow)




git-svn-id: https://svn.xiph.org/trunk/Tremor@17534 0101bb08-14d6-0310-b084-bc0e0c8e3800