An improved version of TSpectrum class. There were two problems.
1. The extrapolation of the right hand part of the spectrum above the edge
(range). Now it works and I do not see any problem.
2. Due to high level of noise in the examples from Sergey the estimation of
the background snathed lower spikes in the spectrum and as a consequence there
remained rests of background after its subtraction. In some cases it was
detected as false peaks. In new version I have included the estimation of
background with simultaneous smoothing, which removes this problem.
However for smooth data and for the detection of small peaks on big background
I recommend to exclude Markov smoothing (set markov parameter to false) when
calling the SearchHighRes function.


git-svn-id: http://root.cern.ch/svn/root/trunk@15368 27541ba8-7e3a-0410-8455-c3a389f83636