Fix panic in fx2

[peterkrack]

I had some issue when calling pineko theory opcards for two of my grids, when the value y = 6.786550974400577 was passed. The newton method here would not converge in 10 steps.

Now using https://crates.io/crates/lambert_w

[cschwan]

1. If you found an example where this function doesn't work, it makes for an excellent regression test. Please add it (to the same file) to make sure that we don't repeat the mistake!
2. In general I recommend to make the minimal changes that fix the problem. In this case especially, because everything is quite sensitive to changes to the x-grid values. Just increase the number of iterations.

[peterkrack]

Changing all the check to be compliant with the solution using the lambert W function is pretty hard. Also the functions converning the compatibility with appl grid files seem to get compromised.

Anyway the issue is related to the FMA Fused-multiply-add-operation. Compiling that function in C or Fortran works only on x86. On arm/M1pro the function will fail for certain values and oscillate if using -ffast-math with -ffp-contract=on. On arm/M1 -ffast-math -ffp-contrac=off the function works. The same oscillating behaviour happens in Rust, which does (and should in general imho) do the FMA optimisation. Maybe there is a way to have just this function unoptimised?

main.c
mainf90.txt

[cschwan]

I believe there's a bug in the Newton's method implementation. The statement delta >= deltap should be: delta.abs() >= delta.abs(). Then the iteration correctly stops, if we also change the limit from 1e-15 to 2e-15.

[cschwan]

This is now ready to be merged. Sorry that I reverted your work, @peterkrack, but I think fixing this bug is the easiest solution.