Set float64

[Giulero]

This PR hopefully fixes "random" failing tests with Jax and Pytorch.
See:

• https://pytorch.org/docs/stable/notes/numerical_accuracy.html
• https://jax.readthedocs.io/en/latest/notebooks/Common_Gotchas_in_JAX.html#double-64bit-precision

This is a preliminary solution, see #42.

[traversaro]

I tried to apply this in the conda-forge PR (see traversaro/staged-recipes@e20e773 and traversaro/staged-recipes@6d839a1) to see if this fixes the failure.

[traversaro]

I tried to apply this in the conda-forge PR (see traversaro/staged-recipes@e20e773 and traversaro/staged-recipes@6d839a1) to see if this fixes the failure.

Apparently the failures continue even with this patch: conda-forge/staged-recipes#21894 and https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=702427&view=logs&jobId=6f142865-96c3-535c-b7ea-873d86b887bd&j=6f142865-96c3-535c-b7ea-873d86b887bd&t=22b0682d-ab9e-55d7-9c79-49f3c3ba4823 .

[Giulero]

That really mysterious 😨
I don't understand. In gravity_term, the only difference with bias_forces is that the robot velocity is zero. What can it be?

[Giulero]

Thanks @traversaro!