Add WithinGibbs sampler implementations

[sunxd3]

This PR:

• Introduces condition and decondition interface similar to those of DynamicPPL, addressing Support condition/decondition syntax similiar to DynamicPPL.  #132
• Implements With-in-gibbs sampler that allow user to specify the variable grouping and sampler(currently support AdvancedHMC, AdvancedMH, and JuliaBUGS.MHFromPrior

[github-actions]

Pull Request Test Coverage Report for Build 6841824404

Warning: This coverage report may be inaccurate.

We've detected an issue with your CI configuration that might affect the accuracy of this pull request's coverage report.
To ensure accuracy in future PRs, please see these guidelines.
A quick fix for this PR: rebase it; your next report should be accurate.

• 70 of 114 (61.4%) changed or added relevant lines in 4 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage decreased (-1.2%) to 74.53%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/graphs.jl	12	13	92.31%
src/markov_blanket_model.jl	50	59	84.75%
src/gibbs.jl	0	34	0.0%

Files with Coverage Reduction	New Missed Lines	%
src/model.jl	1	83.06%

Totals
Change from base Build 6737589416:	-1.2%
Covered Lines:	1349
Relevant Lines:	1810

---

💛 - Coveralls

[codecov]

Codecov Report

Attention: 20 lines in your changes are missing coverage. Please review.

Comparison is base (49d38d4) 75.75% compared to head (120d014) 77.21%.

Files	Patch %	Lines
src/model.jl	82.41%	16 Missing ⚠️
ext/JuliaBUGSAdvancedMHExt.jl	0.00%	3 Missing ⚠️
src/graphs.jl	92.30%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #131      +/-   ##
==========================================
+ Coverage   75.75%   77.21%   +1.46%     
==========================================
  Files          20       21       +1     
  Lines        1757     1826      +69     
==========================================
+ Hits         1331     1410      +79     
+ Misses        426      416      -10     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sunxd3]

@yebai I think this is finally ready. Some tests may fail, but they should be okay, just artifacts from MHFromPrior.

The condition and decondition are added and used.

I removed MarkovBlanketModel, now we only have one kind of model BUGSModel. For models produced by conditioning, their sorted_nodes will only contain the unobserved random variables and corresponding markov blanket instead of all the variables in the model.

[yebai]

The numbers here seem quite off from the ground truth -- maybe double-check the implementation?

[sunxd3]

@yebai I have never seen this happens before, and it's produced by running HMC in a Gibbs fashion. Is it possible that this is just one bad chain?

[sunxd3]

@yebai some tests are passing so I think the implementation is correct. For that I think this PR can be merged.

The issue is that Gibbs runs slowly for many reasons include type unsuitability and unnecessary deepcopies, so I can't make it work and test on slightly larger models. I'll do a refactor and an efficiency update to make it runs better, but that belongs to another PR.