Use JuliaGPU Buildkite CI for GPU tests

[efaulhaber]

Based on #689.

I added a GPU test suite and configured Buildkite on the public JuliaGPU runners.
These resources are very limited, so we should not start a build on every single push. Instead, I added a GitHub action to trigger GPU CI by commenting /run-gpu-tests on a PR. About 30 seconds later, the Buildkite jobs should appear in the list of checks for the PR. Also, every push on main will trigger a GPU CI build.
Note that especially for AMD the lack of resources requires us to wait for about an hour for a free runner.

While writing the GPU tests, I also found the following problems (and fixed some of them):

• Depending on the number of fluid and boundary particles and the number of CPU threads, we sometimes end up in the serial implementation of the Adami pressure extrapolation, which is then still executed in parallel on the GPU, but leads to nondeterministic behavior due to race conditions.
• The RectangularTank setup does not work properly with Float32.
• Pressure clipping in the state equations is hardcoded in Float64.
• EDAC is not GPU-compatible.
• EDAC doesn't work with Float32.
• Moving boundaries are not GPU-compatible.
• The SchoenbergQuarticSplineKernel does not work on GPUs (Can't compare Float32 with Rational on CUDA JuliaGPU/CUDA.jl#2681 and Can't compare Float32 with Rational on Metal JuliaGPU/Metal.jl#550).
• Some weird Heisenbug breaking the 3D dam break on Metal (+= yields incorrect results sometimes JuliaGPU/Metal.jl#549). But only on my machine, apparently.
• The default value for acceleration in the WCSPH constructor is in Float64.
• The open boundary system is not GPU-compatible (Add support for inactive particles PointNeighbors.jl#20).
• Monaghan-Kajtar BC don't work with Float32 (^(::Float32, ::Integer) uses double precision JuliaGPU/Metal.jl#552).
• Intel GPUs just report a segfault in the first example.

[codecov]

Codecov Report

Attention: Patch coverage is 82.08955% with 12 lines in your changes missing coverage. Please review.

Project coverage is 70.25%. Comparing base (aa6475b) to head (49f826c).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/preprocessing/particle_packing/system.jl	0.00%	2 Missing ⚠️
...rc/schemes/fluid/entropically_damped_sph/system.jl	92.30%	2 Missing ⚠️
...rc/schemes/solid/discrete_element_method/system.jl	0.00%	2 Missing ⚠️
src/general/system.jl	0.00%	1 Missing ⚠️
...chemes/boundary/monaghan_kajtar/monaghan_kajtar.jl	66.66%	1 Missing ⚠️
src/schemes/boundary/system.jl	80.00%	1 Missing ⚠️
src/schemes/fluid/transport_velocity.jl	50.00%	1 Missing ⚠️
...s/fluid/weakly_compressible_sph/state_equations.jl	50.00%	1 Missing ⚠️
src/setups/extrude_geometry.jl	85.71%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #723      +/-   ##
==========================================
+ Coverage   70.22%   70.25%   +0.02%     
==========================================
  Files          96       96              
  Lines        6001     6017      +16     
==========================================
+ Hits         4214     4227      +13     
- Misses       1787     1790       +3     

Flag	Coverage Δ
unit	70.25% <82.08%> (+0.02%)	⬆️

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.

[sloede]

With JuliaConcurrent/Atomix.jl#52 merged, at least the oneAPI should be able to run now...

[efaulhaber]

With JuliaConcurrent/Atomix.jl#52 merged, at least the oneAPI should be able to run now...

It's still sh*t and gives a segfault in the first test.

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[svchb]

Highlight support for GPUs by adding it to readme.md