Fixes as suggested in http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/MT2002/emt19937ar.html
to a  known problem in the seeding procedure when a value with many zero in the bit pattern is used (like 2**28).
It is same problem with the 69069 multiplier, which was affecting before TRandom.


git-svn-id: http://root.cern.ch/svn/root/trunk@15438 27541ba8-7e3a-0410-8455-c3a389f83636