np.clip() to avoid floating point round-off errors

Add np.clip() to avoid floating point round-off errors on the macos github CI runners, see also
matplotlib#28823 (comment)

Author: MischaMegens2

lib/mpl_toolkits/mplot3d/axes3d.py

@@ -4053,6 +4053,6 @@ def as_cardan_angles(self):
         qw = self.scalar
         qx, qy, qz = self.vector[..., :]
         azim = np.arctan2(2*(-qw*qz+qx*qy), qw*qw+qx*qx-qy*qy-qz*qz)
-        elev = np.arcsin( 2*( qw*qy+qz*qx)/(qw*qw+qx*qx+qy*qy+qz*qz))  # noqa E201
-        roll = np.arctan2(2*( qw*qx-qy*qz), qw*qw-qx*qx-qy*qy+qz*qz)   # noqa E201
+        elev = np.arcsin(np.clip(2*(qw*qy+qz*qx)/(qw*qw+qx*qx+qy*qy+qz*qz)), -1, 1)
+        roll = np.arctan2(2*(qw*qx-qy*qz), qw*qw-qx*qx-qy*qy+qz*qz)
         return elev, azim, roll