tweak automatic polygon offset check

The current behavior of Gmsh when drawing 3D scalar fields is not coherent
with what is done in 2D: to draw filled iso-values or continuous maps in 3D,
we should ideally do volume rendering. Right now we draw the iso-values
(i.e., surfaces), which is both not coherent with the definition of
continuous maps or filled iso-values, but is also annoying when trying to
visualize fields that are constant per element (since in that case nothing is
displayed!)

Until we can do volume rendering, I've changed the behavior to draw the
solution on the boundary of the 3D elements. It's rather slow, but it's the
right thing to do visually.

small cleanup + don't enter post stuff in select mode at the moment (we
might want to change this if we want to select post-pro views in the future)

instead of drawing in immediate mode (in-between Draw() calls) in HighlightEntity,
just redraw the whole scene. It's slower, but on modern machines it should be OK.
This way we can deal with vertex arrays and mesh highlights easily. I will revert
this change if too many people complain...

- made meshed curves/surfaces selectable with the mouse
- better selection algorithm (use depth info to find the "closest" match)

moving all external code to contrib subdirectory

use it now with -partition N for N partitions
only works with tets and triangles
output should be added (useless for now)

Fixed horrible performance of vertex array creation on Windows for
very large post-pro views (> 1 millon triangles). The bottleneck is
the memory reallocation, which is 100 (?) times slower than on
Linux/Mac. Introduced some simple heuristics to estimate how many
primitives will be displayed. A prepro pass (like for smooth normals)
would be a more elegant solution.

missing cast

fix time step printf format

- new option to choose the mesh element labels (element number, elementary
  entity number, physical entity number or partition number)

finer control of time/timestep display

fix display of normals for very "thin" surfaces with curved boundaries:
this new algorithm should be better

don't draw the trajectory if LineWidth==0

if a curve is "discrete" but has no BDS representation, don't draw anything
in the geometry routine: we draw the elements in the mesh routine.

don't store lines in vol->Simp_Surf : it screws up the saving of Physical
Lines! Use vol->Lin_Surf

fix small bug with overflowing feedback buffer and PDF output

sgi compile fix

update gl2ps

use the partition number and not the parition index to choose the partition
color

increased size of mesh color carousel from 10 to 20

add little menu to choose the quality measure in Options->Mesh->Visibility

fixed small bug when drawing the boundary of a prism for a time step > 1

replaced ExportLCField with ExportMeshStatistics (lc + gamma, eta, rho + elem num)

add note to remember how to use compiled vertex arrays

removed unused stuff + handle SURF_DISCRETE in MeshDiscreteSurface

Fixed performance problem(s) for very large meshes with a huge number of
physical entities:

1) use a temp tree for physical entities in read_mesh

2) we should not set partition=physical for the old mesh format

3) only do getFirstPhysical in draw_mesh if we need to (and warn if
   we do and if there are many many physicals)

optimize 2d iso plots with vertex arrays

polih previous commit

2 big optimizations for meshes that contain a lot of line elements:

- speed up Read_Mesh (by introducing temp tree in curves)
- use vertex arrays for lines in mesh drawing code

update to latest gl2ps