This is going around again, so I'll say this: it's a fine line.
Papers can not have critical technical flaws & still be difficult to follow because the structure is not optimal & arguments are presented confusingly.
Pointing that out is helpful. Knowing when to stop, difficult
As a reviewer, my goal is to find critical flaws, not to "improve" the paper. Every paper can be improved in 100 different ways, many of them subjective. The authors have already decided they're happy with the paper as is. A reviewer is not a co-author or a collaborator.