fix: use global fortran lock instead of per-library

[greglucas]

We are sharing Fortran SAVE state and other global things and using a shared library too, so we need to raise the lock up to the top-level and not try to optimize it for individual libraries. There are likely few people who need the per-library performance benefit.

[greglucas]

Note the real impactful change here was zero-initializing the arrays entering Fortran, it seems like that might not be safe in the Python <-> Fortran bridge layer and some unitialized values get used somewhere only occasionally.