CI: Change actions to self-hosted runner.

[TimoLin]

- The action-runner is running in a docker container hosted on a Win11 computer. It has 12xCPUs and 1x4060Ti GPU card.
- To reduce the ci time, all the prerequisites are installed in the container. This actions script only performs the compiling and testing procedures.

[TimoLin]

Change the runner host from dfgpu container to ubuntu:20.04 container.

In the dfgpu container, the openfoam-7 environment is kind of broken or incompatible with deepflame-dev:

I/root/miniconda3/envs/libcantera/include/python3.8 -I/root/miniconda3/envs/libcantera/lib/python3.8/site-packages/pybind11/include     -I/opt/OpenCC-cuda//include  -DODE_GPU_SOLVER -IlnInclude -I. -I/opt/OpenFOAM/OpenFOAM-7/src/OpenFOAM/lnInclude -I/opt/OpenFOAM/OpenFOAM-7/src/OSspecific/POSIX/lnInclude   -fPIC -c dfLowMachFoam.C -o Make/linux64GccDPInt32Opt/dfLowMachFoam.o
2025-05-19T04:18:04.6589425Z In file included from /opt/OpenFOAM/OpenFOAM-7/src/OpenFOAM/lnInclude/postProcess.H:105,
2025-05-19T04:18:04.6589882Z                  from dfLowMachFoam.C:128:
2025-05-19T04:18:04.6590561Z /opt/OpenFOAM/OpenFOAM-7/applications/solvers/lagrangian/reactingParcelFoam/createFields.H:121:10: fatal error: createRadiationModel.H: No such file or directory
2025-05-19T04:18:04.6591264Z   121 | #include "createRadiationModel.H"
2025-05-19T04:18:04.6591521Z       |          ^~~~~~~~~~~~~~~~~~~~~~~~
2025-05-19T04:18:04.6591771Z compilation terminated.
2025-05-19T04:18:04.6648729Z make: *** [/opt/OpenFOAM/OpenFOAM-7/wmake/rules/General/transform:26: Make/linux64GccDPInt32Opt/dfLowMachFoam.o] Error 1

Using a pure ubuntu:20.04 image will fix this error.
All the prerequisites are installed in the ubuntu:20.04 container, including the flareFGM table file.

[TimoLin]

Current Solution：

• Compiling: 5min
• Test case running
  • Tu500K-Phi1: 39min
  • aachenBomb_2D: 3min
  • dfLowMachFoam/fgm/twoD_SandiaD_flareFGM: Failed
  • dfLowMachFoam/twoD_reactingTGV: 2.5min
  • oneD_detonationH2: 5min
• test: Failed

Todo List:

• In test cases, use GPU to do the inference.
• Use the maximum CPUs to do the parallel simulation.

Suggestion:

Split the compiling and case's tests as two procedures.