Better support the complexity in sample relations

[lambdamoses]

There's a tree, different sections of the the same piece of tissue, different pieces of tissue from the same type of tissue of the same subject, different types of tissues in the same subject, different subjects of the same biological condition, different biological conditions of the same type (like different types of the same disease or treated vs. untreated as opposed to healthy control), different types of biological conditions (like disease vs. healthy). I'm considering trying TreeSummarizedExperiment and see how it goes. It might lead to significant changes in code related to sample_id and a major version bump.

[lambdamoses]

Probably not inheriting from TreeSummarizedExperiment, because I don't need an edge at each cell. The phylo object takes up more memory than a data frame with the sample relations. I only need the tree related sample_id's to say z-planes, subjects, and conditions, so the tips of the tree would be the smallest sample division like z-planes rather than cells. Also, I don't need the edge weights, at least yet.

[lambdamoses]

Actually, I'm not 100% sure if this will go into the Bioc 3.19 release. It might remain a separate branch even if I implement it before the deadline. Because I need to read up and think more about what to do with multiple samples and the sample relations tree. Right now I only have quick and dirty Moran's I comparison between samples, but I think I want to do hierarchical modeling. What this part looks like depends on what I do with it, and it can drastically change when I change my mind about what to do with it. Just as modeling is part of EDA in R for Data Science, it can also be part of ESDA. If it turns out to be a rabbit hole, then I'll just put the quick and dirty thing in the Voyager and put the more refined hierarchical modeling in a separate paper.