Creating a PhysicsSphericalRepresentation with NaNs raises RuntimeWarning #11558

Description

This is not exactly a bug, but inconsistent behavior between the different spherical representations. For example, the following code will raise a RuntimeWarning (and thus error in pytest) if the representation is PhysicsSpherical, but not Spherical.

from astropy.coordinates import PhysicsSphericalRepresentation
from astropy.coordinates.matrix_utilities import rotation_matrix
import astropy.units as u
import numpy as np

s1 = PhysicsSphericalRepresentation(
    phi=[1, 2] * u.deg, theta=[3, 4] * u.deg, r=[5, np.nan] * u.kpc)
matrix = rotation_matrix(-10, "z", u.deg)

# THIS PART RAISES THE WARNING
expected = (s1.to_cartesian().transform(matrix).represent_as(PhysicsSphericalRepresentation))

The reason for this is in the initialization of PhysicsSphericalRepresentation

    if np.any(self._theta < 0.*u.deg) or np.any(self._theta > 180.*u.deg):

I use this example because it fails the test suite in #11444. @mhvk @adrn

Which is not NaN-"protected".

Expected behavior

A NaN-"protected" check of the angular domain.

Actual behavior

Raises RuntimeWarning.

Steps to Reproduce

See decription

Perhaps a simpler example is

from astropy.coordinates import PhysicsSphericalRepresentation
import astropy.units as u
s1 = PhysicsSphericalRepresentation(phi=[1, 2] * u.deg, theta=[3, np.nan] * u.deg)

System Details

macOS-10.16-x86_64-i386-64bit
Python 3.8.2 | packaged by conda-forge | (default, Apr 24 2020, 07:56:27)
[Clang 9.0.1 ]
Numpy 1.18.4
astropy 4.3.dev884+g8c8fe91a6.d20210413
Scipy 1.4.1
Matplotlib 3.4.1