Good suggestion--I should have added the caveat that I set dependencies = FALSE to see if it changed the behaviour. Leaving it as default (dependencies = TRUE) gets the same result but takes a lot longer to run and produces a lot more history output!
This happens from time to time when a new version of dplyr is out, the reason is that dplyr requires more than 1GB of memory to compile (the current RStudio Cloud memory limit), the way RStudio Cloud addresses this issue is by caching precompiled binaries of the packages but it could take some time for the cache to update to the current dplyr version.
That must be the source of the trouble--another thing I failed to mention in my "things I failed to mention" is that the dependencies all installed without issue, so it seems to be specifically dplyr.
I ran into this issue this afternoon while teaching a workshop for students new to R/Rstudio. My students were able to install tidyverse but got an error that dplyr wouldn't install. They also tried installing dplyr on its own, and that also would not work. I think the "base project" solution proposed above won't work for them, since this is their first project. I'll give the workshop again in April, so I will stay tuned for updates