Whenever I'm interested in comparing groups, I usually specify the design for my DESeq2/limma pipeline to include only the variables related to the contrast of interest. Should I include in the design other variables that I'm not interested in contrasting, but to "correct" the model for side effects (kinda the same you do with multivariate survival models)?

Examples could be: RIN scores for my samples or other sources of technical variation, other grouping variables that may be causally related to my comparison of interest.

This is covered in our documentation and I'm sure also in limma docs. Yes, batches should be included in general. Rather than RIN, if there is substantial technical variation, we use RUV or SVA in the lab to estimate continuous nuisance variables.

All the relevant variables should be included in the design matrix. That's a universal principle of any anova/regression analysis in statistics, not just for limma and DESeq2. Otherwise the analysis will be underpowered or biased.

As far as technical corrections go, you should include variables only if they have an meaningful effect for your data. I hardly ever include RIN myself.