* Correct the schema evolution to and from any numerical type involvind on one

  end or the other any variation of Float16_t and Double32_t.
* Add the annotation (aka the comment part with [xyz]) of C style arrays, 
  Float16_t and Double32_t to the checksum calculation.
* Allow the checksum of existing class with a ClassDef to match with and 
  without the annotation taken in consideration.  This  allows for reading 
  older file (hence with desufficient checksum) to be read without any warning 
  message.  However this also means that if the classes on this older file are 
  different _only_ by the annotation, it will not be noted.
* Prevent the fact that any consecutive Double32_t, independently of their
  annotation are 'compiled' together and thus only the annotation of the first 
  one are used (incorrectly).  To support backward compatibility with files 
  written with this bug, the version number of TStreamerInfo has been 
  increased to 6.  The compilation is still done the same way for 'old' 
  TStreamerInfo (TStreamerInfo version 5 or less).  However if the 
  Double32_t are changed to Float16_t, since the compilation is turned off 
  because of the schema evolution, the 'wrong' annotation might be used for 
  those old files. [This should be rare]
* Corrected the fact that due to an off by one lookup, the number of 
  TStreamerInfo for a class was actually limited to 13.


git-svn-id: http://root.cern.ch/svn/root/trunk@20854 27541ba8-7e3a-0410-8455-c3a389f83636