Eliminate flex as the GLSL lexical analyzer, going from two nested lexical analyzers down to one, leaving just the preprocessor's lexical analysis.  A new layer replaces it, to translate from the preprocessor's view of tokenization to glslang's view of tokenization.

Also:
 - change source locations from an int to TSourceLoc (shader number, line number) throughout
 - various improvements to the preprocessor


git-svn-id: https://cvs.khronos.org/svn/repos/ogl/trunk/ecosystem/public/sdk/tools/glslang@22277 e7fa87d3-cd2b-0410-9028-fcbf551c1848