Various minor improvements

[luisenp]

• [ ] Pass dtype to ScaleCostWeight when constructed from float.
• [x] Typo in SE3.
• [x] Add between and adjoint aliases to global th scope.
• [ ] Explicitly disallow non Manifold optimization variables added to Objective.
• [ ] Test old bug in Variable.update(batch_ignore_mask=True) with Lie groups.
• [ ] Consider adding LieGroupTensor wrapper automatically (if needed) when calling TheseusLayer.forward (see).
• [x] Rename pytest.mark.cuda as pytest.mark.cudaext.
• [ ] Make LieGroup.dof() a static method.
• [x] Expose NonlinearOptimizerInfo at th level.
• [ ] Change info.best_iter initialization to a non-ambiguous value (e.g,. -1, currently 0).
• [x] Ensure consistency in the order of args in our cost functions (e.g., optim_vars, aux_vars, cost_weight).