Speedup Slice sampler

[ricardoV94]

Inherits from StepArrayShared to avoid the many mappings between raveled and non-raveled arrays, which greatly slowdown the sampler. This simple model samples 3 times as fast in my machine after the changes:

import arviz as az
import pymc as pm

with pm.Model() as m:
    x = pm.Normal("x", shape=(200,))
    idata = pm.sample(step=pm.Slice([x]), random_seed=1)
    
az.summary(idata)

---

📚 Documentation preview 📚: https://pymc--6711.org.readthedocs.build/en/6711/

[codecov]

Codecov Report

Merging #6711 (eb4becf) into main (a617bf2) will decrease coverage by 31.76%.
The diff coverage is 19.04%.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #6711       +/-   ##
===========================================
- Coverage   89.57%   57.81%   -31.76%     
===========================================
  Files          95       95               
  Lines       16131    16135        +4     
===========================================
- Hits        14449     9329     -5120     
- Misses       1682     6806     +5124     

Impacted Files	Coverage Δ
pymc/step_methods/slicer.py	27.38% <19.04%> (-68.87%)	⬇️

... and 69 files with indirect coverage changes

[fonnesbeck]

Does asarray not make a copy, or is that array?

[ricardoV94]

asarray returns the same object if it's already a numpy array