Refactor rule matching

[ThijsvdLaar]

This PR refactors Message matching for update rule inference. For future rules (e.g. for the planned CVI implementation), the outbound message type might not always be clear upon message rule inference. That is, a CVI node might define Message{FactorNode} as outbound type, even though it really sends a Message{GaussianMeanVariance} upon execution.

The PR introduces << as a "parametric inheritance" operator, such that e.g. Message{Gaussian} << Message{FactorFunction}, but !(Message{FactorFunction} << Message{Gaussian}). The match() function is then defined as a two-way parametric inheritance, such that match(Ma, Mb) = (Ma << Mb) || (Mb << Ma).

This implementation broadens and clarifies match functionality without impacting existing rule matching.

[codecov-commenter]

Codecov Report

Merging #188 (c6bd5e7) into master (fc11951) will decrease coverage by 0.08%.
The diff coverage is 81.53%.

@@            Coverage Diff             @@
##           master     #188      +/-   ##
==========================================
- Coverage   86.59%   86.50%   -0.09%     
==========================================
  Files          95       95              
  Lines        4400     4395       -5     
==========================================
- Hits         3810     3802       -8     
- Misses        590      593       +3     

Impacted Files	Coverage Δ
src/ForneyLab.jl	100.00% <ø> (ø)
src/engines/julia/update_rules/beta.jl	29.03% <33.33%> (ø)
src/probability_distribution.jl	75.34% <33.33%> (-4.37%)	⬇️
src/engines/julia/update_rules/sample_list.jl	100.00% <100.00%> (+10.00%)	⬆️
src/factor_nodes/clamp.jl	94.44% <100.00%> (ø)
src/helpers.jl	62.26% <100.00%> (+1.47%)	⬆️
src/message_passing.jl	84.00% <100.00%> (-0.13%)	⬇️
src/update_rules/beta.jl	100.00% <100.00%> (ø)
src/update_rules/equality.jl	98.09% <100.00%> (ø)
src/update_rules/nonlinear_extended.jl	100.00% <100.00%> (ø)
... and 3 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 fc11951...c6bd5e7. Read the comment docs.

[albertpod]

Looks good to me, however, I will leave the final judgment for merging this to @semihakbayrak.

[semihakbayrak]

Thanks @ThijsvdLaar for "Refactor rule matching" PR. I think it is a necessary extension and I am happy to see this functionality but to be honest I am not that knowledgable about inheritance in Julia and actually encountered several problems before. Nevertheless, as it passed unit tests, I think we can merge it.