Switch PyMC3 to PyMC (which is the new name starting from PyMC v4.0.0)

This MR updates the PyMC3 interface to use the new version of PyMC, which is now v4.0.0 (the version number suffix has been dropped). This is mainly a very straightforward port by replacing imports of pymc3 with pymc. The Theano library has also been renamed to aesara, so those are changed too.

The test suite runs and also the two examples, linear_regression_pymc.py and linear_regression_pymc_custom_likelihood.py, both work exactly as expected.

bilby/core/sampler/ptmcmc.py

@@ -43,9 +43,9 @@ class PTMCMCSampler(MCMCSampler):
 
     default_kwargs = {
         "p0": None,
-        "Niter": 2 * 10 ** 4 + 1,
-        "neff": 10 ** 4,
-        "burn": 5 * 10 ** 3,
+        "Niter": 2 * 10**4 + 1,
+        "neff": 10**4,
+        "burn": 5 * 10**3,
         "verbose": True,
         "ladder": None,
         "Tmin": 1,
@@ -83,8 +83,6 @@ class PTMCMCSampler(MCMCSampler):
         use_ratio=False,
         plot=False,
         skip_import_verification=False,
-        pos0=None,
-        burn_in_fraction=0.25,
         **kwargs,
     ):
 
@@ -99,7 +97,10 @@ class PTMCMCSampler(MCMCSampler):
             **kwargs,
         )
 
-        self.p0 = self.get_random_draw_from_prior()
+        if self.kwargs["p0"] is None:
+            self.p0 = self.get_random_draw_from_prior()
+        else:
+            self.p0 = self.kwargs["p0"]
         self.likelihood = likelihood
         self.priors = priors