Remove max_energy_error sampling stat from HMC

[eigenfoo]

Something I came across while working on LittleMCMC.

Currently it looks like we define the max_energy_error dtype in HMC, but we don't seem to actually have a max_energy_error sampler stat for HMC (although I should note that NUTS does). This PR just removes the max_energy_error dtype.

[codecov]

Codecov Report

Merging #3783 into master will increase coverage by 7.05%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #3783      +/-   ##
==========================================
+ Coverage   83.65%   90.71%   +7.05%     
==========================================
  Files         133      133              
  Lines       20328    20911     +583     
==========================================
+ Hits        17006    18970    +1964     
+ Misses       3322     1941    -1381

Impacted Files	Coverage Δ
pymc3/step_methods/hmc/hmc.py	96.66% <ø> (+3.33%)	⬆️
pymc3/tests/test_ndarray_backend.py	100% <0%> (ø)	⬆️
pymc3/model.py	89.17% <0%> (+0.12%)	⬆️
pymc3/backends/ndarray.py	94.75% <0%> (+1.42%)	⬆️
pymc3/tests/helpers.py	56.6% <0%> (+1.88%)	⬆️
pymc3/sampling.py	85.05% <0%> (+1.93%)	⬆️
pymc3/distributions/shape_utils.py	100% <0%> (+11.11%)	⬆️
pymc3/memoize.py	96.15% <0%> (+13.46%)	⬆️
pymc3/variational/opvi.py	85.29% <0%> (+17.28%)	⬆️
... and 11 more

[ColCarroll]

Thanks, @eigenfoo !