Surface recon determinism

[ClePol]

Summary

This PR contains

1. An improvement of the surface pipeline setting topo fixer to single thread
2. Single threading in spherical project

Testing

End-to-end pipeline run before and after changes on 5 cases.

[m-reuter]

my tests:

• blas on linux gives slightly different outputs depending on the number of threads, so keeping that fixed to one thread is a good idea (also comes at almost no cost as the mesh is relatively small for the solver).
• the env var about numexpr is dead weight as numexpr is not a dependency here. The others don't hurt to keep.
• for the topofixer on a more tricky case I found that 1,1-repeat, 2, 4, 8 and 8-repeat, all give identical results if started from the same initial surface. Speed wise you get 15% improvement going from 1 to 4 threads where it plateaus . So I see no reason to fix it to 1 single thread here.

I will modify the PR accordingly and then merge it. What then remains is to test if the full surface pipeline is now deterministic.