[Tracking issue] gwb-grid improvements

[MFraters]

The following are subprojects to improve gwb-grid:

1. Improve documentation of gwb-grid: Currently there is a short description in the manual here: https://gwb.readthedocs.io/en/latest/user_manual/how_to_use_the_applications/gwb-grid_app.html which can be expanded, but this page: https://gwb.readthedocs.io/en/latest/GWB_parameter_listings/world_builder_file/index.html should also be filled or removed.
2. Improve testing of gwb-grid grids: Currently only cartesian and chunk are tested, but there is also a sphere and an anullus.
3. The gwb-grid should be added to the coverage result: maybe as a separate category, or just lump it in with the rest
4. Better parallelization: gwb-grid currently only has a very simple parallezation scheme. Some kind of job system would be better. (@tjhei?)
5. Add vti output support: this is useful for online visualization in point 7 (@tjhei)
6. Add netcdf output support (@tjhei)
7. Implement online visualization: through glance (https://pvw.kitware.com/glance/) (@tjhei)

[tjhei]

Conversation about better parallelization:

• simple scheduler code based on c++ 11: https://blog.andreiavram.ro/job-scheduler-cpp/
• taskflow https://taskflow.github.io/taskflow/index.html#SupportedCompilers (header-only, but big and requires c++17)

[tjhei]

Do you have an example where the current parallel scheduling is not ideal? I could not reproduce based on your description.

[MFraters]

It mostly shows up at larger models. For example, if you double the resolution of in cookbooks/3d_cartesian_curved_subduction/3d_cartesian_curved_subduction.grid and then run ./bin/gwb-grid -j 6 ../cookbooks/3d_cartesian_curved_subduction/3d_cartesian_curved_subduction.wb ../cookbooks/3d_cartesian_curved_subduction/3d_cartesian_curved_subduction.grid (in release mode), you can see that the first few are done very quickly. The higher resolution, the more the problem creeps up. Doubling the resolution again will make it very clear.

But I noticed that when doing the second doubling, the fact that the creating the grid and the writing of the paraview file are not parallel, is costing way more time at that point.