feat: parallelize structure analysis

[Gitdowski]

After structure analysis has been implemented with a seperate NVT simulation in PR#403 (in case n_averaging_frames > 1) the way how computational resources were used was not correct and inefficient: Because not labelled as a lammps job, the structure analysis only used 1 core. But because mpiexec was used with --oversubscribe the number of cores used for the melt-quenching phase was still used here. Therefore, slurm was submitting to nodes where 1 core was free, but under the hood much more cores were actually used, potentially leading to over-usage of a compute node. This is now fixed:

• Structure analysis is labeled as a lammps job -> receives the same number of cores as the melt-quenching phase for both NVT run and actual structure analysis afterwards
• --oversubscribe is removed -> it should rather throw an error than silently go wild in the future
• after NVT run of structure analysis is finished, the structure analysis is parallelized via new parameter n_jobs
• n_jobs=None is the default, which automatically uses half of the NVT cores. This is a safety measure in case very large structures are processed to avoid memory errors (we can discuss if this is really necessary though, but it shouldn't make such a big difference for the majority of our runs)
• n_jobs is automatically capped at the number of cores used for the lammps
• one of the parallel n_jobs takes care of frame 0 with rings analysis. the remaining n_jobs-1 are used to parallelize the structure analysis of the remaining frames

[codecov]

Codecov Report

❌ Patch coverage is 97.67442% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
amorphouspy_api/src/amorphouspy_api/pipeline.py	85.71%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Atilaac]

thank @Gitdowski for handeling this. I added few comments in the code. that what I could pick for now. once fixed I can approve this.

[Gitdowski]

@Atilaac: Some notebook-invoked simulations for CTE and diffusion failed without a very conclusive error message. It might be that using 4 cores is too much for the github tests. I changed the number of cores to 1 for all these cases, which should not make much of a difference in the overall time for the integration tests. Let's see if this solves the issue.

[Gitdowski]

And thanks for your comments! They should all be fixed now.
I will merge once you have approved the changes and the integration tests run successfully

[Atilaac]

@Gitdowski thank a lot for working on this. I approved the changes and if the notebooks still fail especially the diffusion one I can have a look at the issue and see. Otherwise, feel free to merge.