Substantially Update GMM-01

[JasonPekos]

Ok, this a (fixed) version of #439, ported to Quarto.

Changes from before stay the same. For clarity, these include the following major changes:

• Covers the use of ordered() from Bijectors.jl in making the model identifiable (currently it is multimodal, and the seed is just lucky.)
• Introduces a marginalized version of the model that is much faster.
• Includes a section on recovering parameters from marginalized models

The following minor changes:

• Changed chain plots from grouping chains together to grouping parameters together. I think this makes the multimodality discussion much clearer.
• Added a tiny amount of burn-in to the MCMC sampler defaults, as really far-out initialization was making the plots hard to read.
• Added one more chain to n-chains to make multimodality more likely
• Added and reworked tests for the models (now tests rhat to make sure the bijector discussions are all correct)

There were a few unresolved problems with the last PR (brought up by @devmotion). This PR now deals with those as well:

• The multithreaded calls have the same warning as every other occurrence in the tutorials, so it shouldn't be too confusing now
• The Turing.@AddLogProb! model has a warning under it, telling people to stay away from manually incrementing log probabilities, and pointing to the simplified model underneath so they don't miss it.
• Used LogExpFunctions instead of StatsFuns, moved the assignment-recoverer function out of the model, and just generally changed all the small things pointed out under the last PR.