Split fluid and TLSPH time integration to allow for TLSPH substeps

[efaulhaber]

With high tensile modulus, TLSPH typically requires much smaller time steps than the fluid solver. The vast majority of particles are typically fluid particles, so allowing larger fluid time steps by decoupling the TLSPH time integration greatly improves performance.

Here are benchmarks using RDPK3SpFSAL35 with adaptive time stepping for both the fluid and the solid.
I simulated an elastic plate in a fluid channel in 2D. Flexural rigidity is constant, which means a thinner plate must be stiffer to obtain the same flexural rigidity. A higher resolution is also required for a thinner plate, which drastically increases runtime for thin plates.
The table shows time steps and runtime for 1s of simulation time. The last column shows the percentage of total runtime taken by the TLSPH time integration sub-steps. The benchmarks were run on a 72-core Nvidia Grace ARM CPU.

Thickness	Modulus	Fluid Particles	Solid Particles	Time Steps	Time Steps Splitting	Runtime	Runtime Splitting	Speedup	TLSPH Integration Runtime
10 mm	5.5e7	80k	700	26k	2.5k	640s	82s	7.8×	31%
5 mm	4.4e8	300k	2k	160k	4.3k	3.6h	9.7m	22×	36%
2 mm	6.8e9	2M	10k	2.2M	14k	330h	3.1h	110×	38%
1 mm	5.5e10	8M	35k	14M	31k	400d	34h	280×	46%

I would like to simulate a carbon fiber plate, which has a realistic thickness of 0.5-1mm and a modulus of 1.4e11. It is clear from this table that this will be impossible without split integration (400d vs 34h runtime for a plate that is still too thick).

For such a problem, we get a 300x speedup just with the standard configuration. We can use 450x larger time steps for the fluid in the last row. We can definitely speed this up even more by using specialized time integration for TLSPH, maybe even implicit time integration. This is all trivial with this PR.

[codecov]

Codecov Report

Attention: Patch coverage is 17.12329% with 121 lines in your changes missing coverage. Please review.

Project coverage is 69.28%. Comparing base (785ade6) to head (da28e75).

Files with missing lines	Patch %	Lines
src/callbacks/split_integration.jl	0.00%	114 Missing ⚠️
src/general/semidiscretization.jl	78.57%	6 Missing ⚠️
test/schemes/fluid/rhs.jl	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #794      +/-   ##
==========================================
- Coverage   70.43%   69.28%   -1.16%     
==========================================
  Files         106      107       +1     
  Lines        6867     6996     +129     
==========================================
+ Hits         4837     4847      +10     
- Misses       2030     2149     +119     

Flag	Coverage Δ
unit	69.28% <17.12%> (-1.16%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.