Euler angle functions do not work for float32 arrays

[fbxiang]

Many Euler angle functions produce incorrect results with float32 arrays as input. See the following case.

In [1]: from transforms3d.euler import quat2euler, euler2quat
   ...: import numpy as np
   ...: euler2quat(*quat2euler(np.array([0, 0.70710695, 0, -0.70710695], dtype=np.float32)))
Out[1]: array([0.70710678, 0.        , 0.70710678, 0.        ])

This is reproducible with 0.4.1 and latest main branch.

I believe this error happens because the variable _EPS uses float64 information but float32 arrays are not converted to float64 before computation.