IMHO the syntax & other nitpickable matters should be caught by the linter. The regressions should be caught by the tests. And the fact the code runs and works should be a given. The discussion should be around the architecture of the solution & the long terms consequences of solving it that way. I also like to do the reviews in person, where we just go over the code together and talk about it. Small remarks can be immediately fixed

Also I think work in progress should be pushed often in a wip PR/MR, that way people can give early feedback