Bug in ppc conda package

Found using the conda installation on Satori. The log-determinant of affine maps is wrong. It seems like the value of the determinant is getting rounded at some point (note: I changed the python test to compare the determinant here, not the log-determinant).

E       AssertionError: assert False
E        +  where False = <function allclose at 0x7ff759de30d0>(array([7., 7., 7., 7., 7., 7., 7., 7., 7., 7.]), 7.500000000000003, atol=1e-08)
E        +    where <function allclose at 0x7ff759de30d0> = np.allclose
E        +    and   array([7., 7., 7., 7., 7., 7., 7., 7., 7., 7.]) = <ufunc 'exp'>(array([1.94591015, 1.94591015, 1.94591015, 1.94591015, 1.94591015,\n       1.94591015, 1.94591015, 1.94591015, 1.94591015, 1.94591015]))
E        +      where <ufunc 'exp'> = np.exp
E        +    and   7.500000000000003 = <function det at 0x7ff759c895e0>(array([[ 1. ,  2. ],\n       [-3. ,  1.5]]))
E        +      where <function det at 0x7ff759c895e0> = <module 'numpy.linalg' from '/nobackup/users/mcbrenn/.conda/envs/mpart-conda/lib/python3.9/site-packages/numpy/linalg/__init__.py'>.det
E        +        where <module 'numpy.linalg' from '/nobackup/users/mcbrenn/.conda/envs/mpart-conda/lib/python3.9/site-packages/numpy/linalg/__init__.py'> = np.linalg

Reproduce the failure from the modified Python test described in the issue using the conda installation on Satori, then trace the affine-map log-determinant implementation across the Python and C++ package layers. The fix is done when the determinant for the shown matrix is preserved as 7.5 rather than rounded to 7 and the regression test passes.