Add HMM-constructing helper functions and Ops

[brandonwillard]

This PR adds HMM constructor functions and an OpFromGraph example for a discrete Markov chain. The latter demonstrates how "composite" measurable Ops can be used in AePPL, and the former shows how simple constructor functions that make use of our automatic log-probability derivations can be used to construct interfaces for more advanced models.

Once #75 goes through, we can remove the composite Op approach involving DiscreteMarkovChainFactor, but—in the meantime—it serves as a good example.

Also, these additions will complement an upcoming implementation of aesara-devs/aemcmc#6.

[ricardoV94]

Quite interesting. Did you find any issues with scan + rng?

[brandonwillard]

Quite interesting. Did you find any issues with scan + rng?

Scan can be used to generate a log-probability for models like this right now, but, without #75 in place, the stochastic state sequence inputs can't be defined anywhere except inside the Scan.

[codecov]

Codecov Report

Merging #130 (441ce60) into main (a97a6bc) will increase coverage by 94.27%.
The diff coverage is 93.42%.

@@            Coverage Diff            @@
##           main     #130       +/-   ##
=========================================
+ Coverage      0   94.27%   +94.27%     
=========================================
  Files         0       11       +11     
  Lines         0     1606     +1606     
  Branches      0      228      +228     
=========================================
+ Hits          0     1514     +1514     
- Misses        0       51       +51     
- Partials      0       41       +41     

Impacted Files	Coverage Δ
aeppl/printing.py	89.68% <ø> (ø)
aeppl/dists.py	94.11% <91.93%> (ø)
aeppl/logprob.py	98.80% <100.00%> (ø)
aeppl/transforms.py	94.49% <0.00%> (ø)
aeppl/abstract.py	100.00% <0.00%> (ø)
aeppl/truncation.py	98.43% <0.00%> (ø)
aeppl/mixture.py	98.02% <0.00%> (ø)
aeppl/opt.py	81.41% <0.00%> (ø)
... and 4 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 a97a6bc...441ce60. Read the comment docs.