Allow pytorch batching using jax2torch

[Giulero]

This PR exploits the package jax2torch, based on the gist which explains how to convert jax functions in pytorch ones, preserving also the gradients.

This should allow to perform batch computations, e.g.

H_b_batch = torch.tile(torch.tensor(H_b), (n_samples, 1, 1)).requires_grad_()
joints_val_batch = torch.tile(torch.tensor(joints_val), (n_samples, 1)).requires_grad_()

mass_matrix = kindyn.mass_matrix(H_b_batch, joints_val_batch)
# The mass matrix dimension should have dimension (n_samples, n_dofs + 6, n_dofs + 6)

# and gradient
mass_matrix.sum().backward()

I put this interface in adam.pytorch as KinDynComputionBatch, even if it using jax under the hood.

from adam.pytorch import KinDynComputationsBatch
# and it is instantiated in the vary same way of the other interfaces
kindyn = KinDynComputationsBatch(model_path, joints_name_list)

@traversaro already opened conda-forge/staged-recipes#26780 (comment), so everything will be soon conda ready.

[traversaro]

@traversaro already opened conda-forge/staged-recipes#26780 (comment), so everything will be soon conda ready.

Merged: https://github.com/conda-forge/jax2torch-feedstock .

[Giulero]

All tests are passing now!

[Giulero]

Cc @evelyd @Zweisteine96