Parallelization

[carloalbert]

1. Update every particle in population in parallel
2. Updates of epsilons and Sigma only after each population update
3. Each thread counts number of accepted updates
4. If total accepted updates after one population update larger than x% of n_particles, do the updates of epsilon and Sigma
5. If total accepted updates after several pop updates larger than n_particles, do a importance sampling

[scheidan]

We should read this blog carefully: https://julialang.org/blog/2023/07/PSA-dont-use-threadid/

I do completely understand the problem they are pointing out yet.

See also this discussion: https://discourse.julialang.org/t/psa-thread-local-state-is-no-longer-recommended-common-misconceptions-about-threadid-and-nthreads/101274