add low-mach correction to HLLC solver

[BenWibking]

This will be useful for many galaxy cluster applications.

There is

• HLLC-LM: https://ui.adsabs.harvard.edu/abs/2020JCoPh.42309762F/abstract
• low-dissipation HLLD: https://ui.adsabs.harvard.edu/abs/2021arXiv210804991M/abstract

HLLC-LM is not stable for very low-amplitude linear waves, but it works well otherwise. I'm not sure which is best. Both have ad-hoc modifications to recover the $\mathcal{M} \rightarrow 0$ limit, and both suppress carbuncles.

[pgrete]

If you quickly want to try it, I have a hacked version of the latter in a branch for testing (main...pgrete/lhlld)
Didn't fix my cloud issues but I should probably nevertheless push this upstream at some point.

[pgrete]

For reference on how to port Athena++ solvers to AthenaPK: https://gitlab.com/theias/hpc/jmstone/athena-parthenon/athenapk/-/merge_requests/32/diffs?commit_id=258905f9baadef3e1cca6f977fff79dac1ee2b43 for updating the implemention itself and https://gitlab.com/theias/hpc/jmstone/athena-parthenon/athenapk/-/merge_requests/32/diffs?commit_id=6947bc33a33e038a8f6d2d83ae81f240387e7f2e to integrate the riemann solver

[BenWibking]

The main thing complicating this appears to be the addition of a separate ComputeVelocityDifferences kernel.

In Quokka, we found that computing the velocity differences inside the Riemann solver kernel itself improved performance slightly. The code is here, but there is some rather inelegant index manipulation involved: https://github.com/quokka-astro/quokka/blob/4c34ac8fd1299192ef8934b7959eb73f83b27bea/src/hydro_system.hpp#L897. I can probably port this over after the ASU conference.