Consider getting rid of the _pre_warmup and _pre_sample methods in NUTS

[amal-ghamdi]

Description:

consider getting rid of the _pre_warmup and _pre_sample methods in NUTS. See the discussion in #408. This will also lead to updating the new Sampler class and HybridGibbs to remove special treatment for samplers that have pre_sample and pre_warmup

DoD:

• investigate whether this (removing the two methods) is feasible
• If removing them is feasible, remove _pre_warmup and _pre_sample from all classes mentioned above
• Add regression test for NUTS within Gibbs to make sure as of this update the results are consistent with the main branch
• The PR is reviewed and merged

[amal-ghamdi]

Can you review the DoD, @jakobsj ? Thank you.

[jakobsj]

@amal-ghamdi any unit tests to modify or remove? If not then approved

[amal-ghamdi]

Thank you @jakobsj , Yes good point, I added a test for NUTS within Gibbs in DoD.

A related point, I also investigated the already-existing regression tests (from when we made the new NUTS in the experimental module) to make sure chains are actually identical (between old and new NUTS), and it was. So I think it is robust in terms of matching the old NUTS.