feat!: generalize away condition

[dlaehnemann]

This pull request mainly generalizes the configuration of DESeq2 differential expression analysis in a way that closes #53. Every use of condition (except for one very basic test case) is changed to variable_of_interest in general settings and to specific variable names wherever something is configured for a particular analysis. The aim is to always be explicit about configuration, as opposed to relying on defaults in DESeq2 (which I find confusing at times).

The new setup allows for:

• Multiple variables_of_interest. For each of them, a base_level needs to be explicitly specified and is set up accordingly in deseq-init.R using relevel() (and also ensuring that all of these variables are turned into factor()s, as this would not happen if all entries in a column are numeric). This explicit specification ensures that users get the fold change that they are actually looking for. If multiple variables_of_interest are specified, the model will per default include all the possible interactions, so the resulting formula will contain something like vof1 * vof2 for two variables_of_interest of that name. If you wonder why I chose this as a default, there's a nice and concise demonstration on why you want the interaction * to be the default over the addition + of terms. But if the user is sure that they actually don't want interaction terms, they can override this by manually providing a formula under model: "".
• Multiple batch_effects, which are added to the model with + (to exclude their effect from the other terms).
• Multiple contrasts. These each trigger a separate call of deseq2.R, namely the results() function of DESeq2. For each of the contrasts: specified, two out of three options that the ?results help page describes for the contrast argument can be used:
  1. Specifying a variable_of_interest and the level_of_interest will result in a contrast c(variable_of_interest, level_of_interest, base_level).
  2. Specifying a string can be used to provide a list of (one or) two character vectors that can contain all the resultsNames(dds_object), for example something like: "list(c("genotypeIII.conditionB"), c("genotypeII.conditionB"))'.

In addition, I fixed some other things in this pull request, as this will mean a major version bump, anyways:

• applied some R lints to the R scripts
• some cleanup of comments
• included a more complex testing setup needed for the variables_of_interest changes
• fixes rule align (star) parameters and make gtf an actual input
• reworked the entire config/README.md to better explain stuff for users via the snakemake workflow catalog

[johanneskoester]

Very nice work!

[jonathandmoore]

Thanks for inviting review, looks like I missed the merge, but great to see the improvement.