second order

reduce dependencies by putting the selection buffer stuff in its own file

small cleanups + start rewriting 2nd order stuff

start removal of unused stuff

only compute mesh quality measures on demand: it's too slow for large meshes

Implemented new binary MSH file format (both read and write).

A 250Mb mesh file (6 million tetrahedra) is now read or written to disk in
less than 10 seconds (on my powerbook's slow hard drive).

added binary stl export

- rewrote selection code to use dynamic arrays, so that we should never
overflow the selection buffer again (even on huuuuge models)
- cleanup some leftovers "useing namespace std"

changed default hover+selection mode as follows:

- by default, mouse hover+selection is ON for the geometry, but only
  mouse selection is ON for the mesh

- with Shift+Escape, one can activate "full" selection mode, i.e., also
  allow mouse hover for the mesh

I've disabled the hover over mesh stuff due to slow performance on both
of the Windows machines I have access to over here.

With these new settings I get interactive performance (about 4 frames
per second) when visualization a 6 million tetrahedra mesh, with all the
edges inside visible, lights turned on, etc. This takes about 600Mb of
RAM.

better fix: *only* start from scratch if we actually deleted something

make sure to rest the model before doing a add_in_file in the old geo code
(since the parsed command could delete an entity!)

let's add elements after existing ones in regions (instead of copying the
whole vector, overwriting preexisting data); this way we can easily merge
multiple meshes (with duplicate nodes)

enable multiple de-selections at once using the lasso

small simplification + color bug fix

- added 'Delete' button in visibility browser so that we can easily remove
  physicals and partitions from a model

- slightly changed the behaviour of "Save all (discard physicals)" in the msh
  format: now we really discard the physicals, and set the physical-region
  tag to 0 in the msh file. This is the most natural thing to do: this
  way when on reads a msh file, we get the correct information back.

new option to reverse all the normals (for display). Man, the new code rocks :-)

set light on for mesh by default

testing new default color scheme

fix import of VRML polys with numvert > 4

The reorganisation of the option panel and the new mesh I/O layer are done.

Mesh I/Os currently available:

- Gmsh MSH format: read and write
- Gmsh POS format for mesh statistics: read and write
- INRIA MESH format: read and write
- STL: read (ascii and binary) and write
- VRML: read (partial) and write
- UNV: write

Everything is significantly faster than before (e.g. readMSH() can be more
than twice as fast when there are lots of physicals)

The rewrite of the geometry and mesh drawing code is complete.

There is still room for optimizations, but mostly at the level of the
algorithms (edge/face generation, striping, etc.)---changes that will
be easy do do thanks to the new much cleaner interfaces.

From preliminary tests (even without new algorithms) the code seems to
be faster (a LOT faster for volume meshes) and to use quite a bit less
memory.

I'm tagging the trunk as "gmsh_2_0_alpha_1".