Refactor jax internals to support dense_mass kwarg for numpyro

[ferrine]

What is this PR about?
Enables Block Dense mass matrix adaptation for numpyro

Checklist

• Explain important implementation details 👆
• Make sure that the pre-commit linting/style checks pass.
• Link relevant issues (preferably in nice commit messages)
• Are the changes covered by tests and docstrings?
• Fill out the short summary sections 👇

Major / Breaking Changes

• ...

New features

• Block mass matrix for numpyro
• get_jaxified_logp now accepts point_fn argument

with pm.Model(
        coords=dict(level=["Basement", "Floor"], county=[1, 2]),
) as model:
    # multilevel modelling
    a = pm.Normal("a")
    s = pm.HalfNormal("s")
    a_g = pm.Normal("a_g", a, s, dims="level")
    s_g = pm.HalfNormal("s_g")
    a_ig = pm.Normal("a_ig", a_g, s_g, dims=("county", "level"))
    trace = sample_numpyro_nuts(
        nuts_kwargs=dict(
            dense_mass=[
                ("a", "a_g"),
            ]
        )
    )

Bugfixes

• ...

Documentation

• ...

Maintenance

• ...

---

📚 Documentation preview 📚: https://pymc--7050.org.readthedocs.build/en/7050/

[codecov]

Codecov Report

Merging #7050 (8eb4284) into main (2e05854) will decrease coverage by 12.23%.
The diff coverage is 0.00%.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #7050       +/-   ##
===========================================
- Coverage   92.19%   79.97%   -12.23%     
===========================================
  Files         101      101               
  Lines       16893    16911       +18     
===========================================
- Hits        15575    13524     -2051     
- Misses       1318     3387     +2069     

Files	Coverage Δ
pymc/sampling/jax.py	0.00% <0.00%> (-93.08%)	⬇️

... and 31 files with indirect coverage changes

[ricardoV94]

These failing tests are definitely a latest PyTensor issue, I'll patch it

[ricardoV94]

Failing tests due to PyTensor should be fixed by pymc-devs/pytensor#546

[ricardoV94]

@ferrine can you rebase?