Looking at journal articles, there are two broad approaches to single-cell data preprocessing; Bioconductor-centric and Seurat-centric. The first typically uses log-normalised counts and the second typically uses SCTransform followed by centring and scaling, each followed by their own conventional clustering variety. Is there any direct comparison of these two alternatives, perhaps on simulated data with a known truth?

It's Bioconductor, Seurat or ScanPy to be precise that are most commonly used.

There is benchmark literature on all sorts of preprocessing steps, for example transformations: https://pubmed.ncbi.nlm.nih.gov/37037999/