GitHub actions: test case `Tu500K-Phi1` takes too much time.

[TimoLin]

Tu500K-Phi1: a one-dimensional freely propagation flame.

• Case path: https://github.com/deepmodeling/deepflame-dev/tree/master/test/Tu500K-Phi1
• mesh: 880
• simulation: 0~0.002s (dt=10^-6s)
• DNN inference: on CPU

Using 4 cpus, it takes about 30~40minutes to complete the simulation. Usually, it only takes a few minutes.

Log of the action:

2025-05-19T07:13:29.4576504Z Running blockMesh on /root/actions-runner/_work/deepflame-dev/deepflame-dev/test/Tu500K-Phi1
2025-05-19T07:13:29.4900159Z Running decomposePar on /root/actions-runner/_work/deepflame-dev/deepflame-dev/test/Tu500K-Phi1
2025-05-19T07:13:29.5397954Z Running mpirun on /root/actions-runner/_work/deepflame-dev/deepflame-dev/test/Tu500K-Phi1
2025-05-19T07:52:14.9062881Z /*---------------------------------------------------------------------------*\
2025-05-19T07:52:14.9063260Z   =========                 |
2025-05-19T07:52:14.9063462Z   \\      /  F ield         | OpenFOAM: The Open Source CFD Toolbox
2025-05-19T07:52:14.9063792Z    \\    /   O peration     | Website:  https://openfoam.org
2025-05-19T07:52:14.9064292Z     \\  /    A nd           | Version:  7
2025-05-19T07:52:14.9064462Z      \\/     M anipulation  |

Potential solutions:

• The network is too huge as a test case.
• Use GPU instead of CPU to do the inference.

[TimoLin]

Fixed in #557 by replacing the network using a smaller one. Fix patch is provided by @xu-23