Speed up computation of contact jacobians

[diegoferigo]

The computation of the Jacobians corresponding to the implicit frames associated to the collidable points was introduced in #163. At that time, there was a difference in the body frame considered in the Jacobians computed with jaxsim.api.model.generalized_free_floating_jacobian, that was fixed in #167.

For this reason, originally the code to compute the contact Jacobian was calling jaxsim.link.jacobian inside a jax.vmap transformation. This PR modifies this logic to compute the Jacobians of all links only once, and then with jax.vmap extract the one corresponding to the link to which the collidable point is attached.

On the ErgoCub model, that has 32 collidable points, the speedup is significant:

# Before
8.46 ms ± 574 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

# After
262 µs ± 6.87 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

---

📚 Documentation preview 📚: https://jaxsim--188.org.readthedocs.build//188/

[flferretti]

Great, thanks Diego! LGTM