Onelab "changed" flag is now multi-level (not just true/false). This allows finer granularity for complex clients. For example, Gmsh now responds to 4 levels of "changed"

- changed == 0: nothing to do
- changed == 1: only save the mesh (useful if we only chanaged some physical groups)
- changed == 2: only remesh and save the mesh (useful if we only changed some meshing paramaters)
- changed >= 3: reload geometry, remesh and save the mesh

- removed onelab::region and onelab::function (they complexify the interface, but bring no additional value compared to onelab::string)
- removed ONELAB2 experiment
- made client/Model name persistent, and define Gmsh/Model name so that we can uncouple e.g. .geo/.pro pairs

* New ENABLE_COMPRESSED_IO option to enable on-the-fly IO with gzipped files (patch from Alexis Salzman)
* New PrintParserSymbols() to print workspace (currently in option window, for debugging puroposes)

Fopen deals with non-ASCII filename paths on Windows.

- fixed MED node ordering
- polished 64 bit Cocoa Mac OS X FLTK version
- updated copyright

--no commit message

Propose to automatically create .geo file with a Merge command when
trying to append a scripting command to a non-geo file. Got the idea 
after the discussion with Robert Bou. (This is better than scripting
the Merge command IMHO.)

- invert #define logic (HAVE_MESH, HAVE_POST, HAVE_SOLVER, ... instead of HAVE_NO_MESH,
  HAVE_NO_POST, HAVE_NO_SOLVER)

re-committing work from last two days after server crash:

- refactored old Geo code (new/delete)
- moved DrawPlugin to Plugin
- move from matheval to mathex
- move from Flu_Tree_Browser to Fl_Tree
- fixed crash in Triangulate plugin

- moved GUI to FlGui + added experimental QGui to test Qt

- fixed most MSVC and gcc compiler warnings

- converted some remaining tabs to whitespace (that's why the patch is
  so large; please try to remember to indent with 2 spaces and not
  tabs when you commit stuff to CVS)

- merged modified patch from Konstantinos Poulios to allow saving
  physical/partition as element tags in mesh formats that don't
  support them natively

HAVE_XXX defines.

This fixes three longstanding problems:

- the API was hard to use because some headers had HAVE_XXX ifdefs, so
  that cient codes had to know how the lib was compiled to compile
  themselves

- we had to do make clean after each ./configure, which erased too
  much (e.g. most of contrib)

- make depend could include unwanted stuff since it redefined FLAGS