Improve mesh generation

[perrette]

Currently the mesh is define from three skeleton lines (middle, left, right, downstream looking), by drawing orthogonal segments to the central line at regular intervals, and finding the intersection points with the side lines. This is a very crude approach which has a number of inconvenient:

• the central line should be smooth enough, otherwise the mesh will be seriously distorted
• a numerical integration over the resulting mesh does not ensure conservation of mass, or whatever physically meaningful quantity
• the central line should actually be enough to generate the mesh...

Ideas for solutions

• introduce a GUI option to smooth the lines before generating the mesh, say using second-order splines with 3 or 5 neighbors.
• looking into better meshing algorithms...
  • In particular, it would be neat to just have to define the glacier mouth (cross-flow) and have the mesh generated from there. The algorithm should involve a smoothing or correction part for the velocity field (e.g. making it divergence free), so that it is not too sensitive to errors.
  • do not need straight lines as cross-sections. Flow convergence at narrowing sections should result in slightly bent cross-sections with middle point pointing upstream.