I guess they just mention the strategy. A lot of papers don't follow just one algorithm. For example, Seurat for clustering ,edgeR for DGE and so on. A lot of early papers didn't even use Seurat. Even when people use Seurat, they just say we used Seurat under default conditions. However, genomic studies comparing multiple algorithms do submit there code online via github.

In general, no, though I'd also love to see reproducible code as a requirement, particularly for single cell data.

Typically this is not done, but would be desirable. Using something with default settings might sound trivial to a wetlab review and is therefore often tolerated but every experienced analyst knows that you do not simply press the Run button somewhere in order to get results. You always have to put together at least some custom code. If a benchmarking paper does not provide code I do not even read it anymore. It simply has no use without exact parameters they used.