Use libosmesa for headless rendering in Colab

[flferretti]

This simplifies the headless rendering in Colab, avoiding both to set an additional environment variable and to install pyvirtualdisplay in combination with xvfb and egl.

---

📚 Documentation preview 📚: https://jaxsim--214.org.readthedocs.build//214/

[diegoferigo]

It's probably worth also taking advantage of this PR to simplify the positioning of the recording camera with something similar to the snippet below:

mjcf_string, assets = UrdfToMjcf.convert(
    urdf=model.built_from,
    cameras=jaxsim.mujoco.loaders.MujocoCamera.build_from_target_view(
            name="recorder", lookat=data.base_position(), distance=3, azimut=180, elevation=-20
    ),
)

and then recorder.record_frame(camera_name="recorder").

[flferretti]

Added example in 5123e78. Ready for review, thanks!

[diegoferigo]

Great, thanks!