If we want to start nit-picking the RFC (7230), Connection header follows Connection = 1#connection-option, which means it’s a comma-separated list and fits the exception in your quoted part.
As far as the extra space in the transfer-encoding header goes, it is weird and inconsistent. It does go against the preference set in the RFC, but it is allowed in the RFC. Should it, in a perfect world, be fixed for consistency? Yes. Is it ever going to be the highest priority item on the todo list? No, if v5 is going to go away as planned.