From Lorenzo:

patch for fixing the following problem with BayesDivide in TGraphAsymmErrors:

 - missing precision when inverting the beta distribution (see bug https://savannah.cern.ch/bugs/?30246

- use lower(upper) error equal zero for bins with  0 (1) efficiency


git-svn-id: http://root.cern.ch/svn/root/trunk@21323 27541ba8-7e3a-0410-8455-c3a389f83636