Tests based on panda3d bug when chosing the device

[ElliotMaitre]

When running test on a device with cuda available, we want to test that it both works on GPU and CPU. In this case, different problem arise when running the tests. Problematic tests are test_renderer_panda3d.py and test_megapose_inference.py.

In order to make the test run on a chosen device, I tried to modify some functions in the following way :

load_named_model, load_pose_models, Panda3dBatchRenderer, Panda3dSceneRenderer, App (i.e. Panda3d App) and the worker_loop of Panda3dBatchRenderer take device as input. Then, the different lines involving code such as if torch.cuda.is_available() is changed to use the selected device. This change seems to cause some issues :

• If any test involving pybullet is run before a test involving panda3d, the test bug and a Python aborted error occurs. The current workaround is to change the test order, since the usage of both pybullet and panda3d in the same run is currently not possible, except in the test environment. In this setup, test_megapose_inference.py runs
• Panda3dSceneRenderer related tests are passing. Panda3dBatchRenderer suffers a Python Aborted error. The issue arise in _init_renderer, line 331 renderer_process.start()

For now, we decided to test only on GPU if a GPU is available, and on CPU if not.