Allow sampling in aligned spin and spin magnitude

This MR will allow us to sample in the spin magnitude and the aligned component of the spin, rather than magnitude and tilt.

The reason to do this is that the aligned component of the spin is the best measured per component spin parameter. From preliminary tests, it looks like this will improve convergence with at least MultiNest.

Currently, this will only support the vanilla flat in magnitude, isotropic in orientation prior. It would take more effort to figure out the generic case.

This required making the CBCPriorDict a subclass of the ConditionalPriorDict, I know that people have reported seeing issues with this in the past, but I don't remember what they were. @moritz.huebner @rory-smith?

EDIT I think the issue with the conditional priors was the _check_if_prior_can_be_sampled method which I have removed. That was falling over for some conditions/constraints. I think the current version should be suitably robust.

test/core/prior/prior_test.py

@@ -667,17 +667,6 @@ class TestPriorClasses(unittest.TestCase):
         prior_2.other_key = 5
         self.assertNotEqual(prior_1, prior_2)
 
-    def test_np_array_eq(self):
-        prior_1 = bilby.core.prior.PowerLaw(
-            name="test", unit="unit", alpha=0, minimum=0, maximum=1
-        )
-        prior_2 = bilby.core.prior.PowerLaw(
-            name="test", unit="unit", alpha=0, minimum=0, maximum=1
-        )
-        prior_1.array_attribute = np.array([1, 2, 3])
-        prior_2.array_attribute = np.array([2, 2, 3])
-        self.assertNotEqual(prior_1, prior_2)
-
     def test_repr(self):
         for prior in self.priors:
             if isinstance(prior, bilby.core.prior.Interped):