Fix performance for parallel sampling with mc backend

Somewhere during the release of 1.2.0 a bug snuck in, which makes parallel calls to hopsy take twice as long as they should.

When we tackle this problem, we should also make sure, that parallel sampling works with PyProposals, see examples/BenchmarkingMultiphaseMonteCarloSampling.ipynb

See attached plots

The performance graph we should see for 16 cores and 16 chains:

The performance graph we see for 16 cores and 16 chains:

Benchmark command was

psrecord "python examples/oseir_mcmc.py"  --interval 1 --plot performance_plot_hopsy1.3.1.png --log activity.txt --include-children

Ask me for more details.