Fix rotation conversion signs at Euler singularities

[YaoxinHuang]

At gimbal lock, rmat_to_euler can return angles describing a different rotation for the ZYX, XZY, and YZX orderings. For example, converting [0.3, -pi/2, -0.7] in ZYX order to a matrix and back introduces a 180-degree rotation error.

Correct the signs passed to arctan2 in the affected pole branches. Add rotation-matrix round-trip tests covering both poles and nonzero outer angles for all six Tait-Bryan orderings; the tests compare orientations because Euler angles are not unique at gimbal lock.

Validation: python -m pytest dm_control/utils/transformations_test.py -q.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.