Msm submodels

[clonker]

Refactored MSMs so that the count model can create submodels, there is no more active set and no pre-selection of the largest connected set.
A consequence of this is that the ML-MSM estimator takes either an estimated count model (ie with dtraj statistics) or a count matrix for estimation.

Count models require only a count matrix to be constructed and MSMs only require a transition matrix to be constructed.

This PR is currently failing and WIP because of the lack of a consistent implementation in the HMSM case.

[codecov]

Codecov Report

Merging #39 into master will decrease coverage by 0.29%.
The diff coverage is 90.68%.

@@            Coverage Diff            @@
##           master      #39     +/-   ##
=========================================
- Coverage   83.97%   83.67%   -0.3%     
=========================================
  Files          69       68      -1     
  Lines        4704     4620     -84     
=========================================
- Hits         3950     3866     -84     
  Misses        754      754

Impacted Files	Coverage Δ
sktime/markovprocess/koopman_reweighted_msm.py	39.39% <0%> (ø)	⬆️
sktime/markovprocess/bayesian_msm.py	100% <100%> (ø)	⬆️
sktime/clustering/kmeans.py	93.51% <100%> (ø)	⬆️
...ovprocess/bhmm/estimators/_tmatrix_disconnected.py	86.8% <100%> (ø)	⬆️
sktime/markovprocess/bhmm/init/discrete.py	92.92% <100%> (ø)	⬆️
sktime/markovprocess/__init__.py	100% <100%> (ø)	⬆️
sktime/markovprocess/util.py	95.45% <100%> (+1.7%)	⬆️
sktime/base.py	92.24% <100%> (-0.49%)	⬇️
sktime/markovprocess/reactive_flux.py	93.28% <100%> (ø)	⬆️
sktime/markovprocess/pcca.py	98.21% <100%> (ø)	⬆️
... and 17 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6decbdf...a574349. Read the comment docs.