Fix error in Categorical prior and add tests

In the initial implementation, the rescale() function was biased and did not evenly sample from all categories (the first and last categories where downweighted). This fixes the issue and adds a test.

bilby/core/prior/analytical.py

@@ -1469,7 +1469,7 @@ class Categorical(Prior):
         =======
         Union[float, array_like]: Rescaled probability
         """
-        return np.round(val * self.maximum)
+        return np.floor(val * (1 + self.maximum))
 
     def prob(self, val):
         """Return the prior probability of val.