Is boosting and bagging only relevant in the context of decision trees?

So bagging and boosting are techniques, these two are aimed at slightly different things. Bagging is aiming to reduce variance of the predictions by creating an ensemble of multiple models. Boosting is aiming at decreasing the bias of the predictions by focusing better on data instances that the previously trained model in the ensemble incorrectly classified.

Now on to your major part of the question. Realistically bagging can be used on any model to create an ensemble. As it focuses on reduction of variance, decision trees which tend to over fit when left with no maxdepth is a good model to use for explanation context. However, you could use Logistic regression, or Neural Networks. So long as the bagging approach is followed you'll get the same results. Note: Logistic Regression really doesn't need it, as they tend to not over fit.

Boosting is again just easier to discuss when talking about decision stumps (1 depth decision trees) as they are simple high bias algorithms which boosting aims at and is good at improving.

Boosting and bagging can be applied to various types of classifiers. When a classifier estimates posterior class probabilities, combining several of such classifiers is straight forward. Two classsic references are:

Kittler on combining classifiers and

Hansen on ensembles of neural networks