I do not know if this is indeed the solution, but `evil-yank' is in the end a usual Emacs command, i.e., it has an interactive-form. Your redefinition does not have an interactive form and is therefore not regarded as a command. Changing your code to
... (flet ((evil-yank (&rest args) (interactive) t)) ...
may work. But I haven't tested it.
Thank you very much for the help. Adding "(interactive)" didn't work, and even worse, previously I could re-evaluate the `evil-yank' (C-M-x) to make yank work again but after adding "(interactive)", "C-M-x" no longer do the trick and I have to restart Emacs to make "yank" work again, provided that I don't delete blank lines.
Furthermore it may be easier to just call `delete-region' to remove those lines instead of redefining `evil-yank'. In fact, this is what `evil-delete' does.
Yes, that would be easier and I may have to go that way finally if I can't make this work. However, in general, I think it is still better not to bypass the the original function altogether ('evil-delete' here) so that I will still be able to take advantage of new changes in the original function without having to monitor possible changes in the original function and port them into my code in the future. But yes, in this particular situation the `evil-delete' may not be changed anymore and your suggestion might be better. However, if possible I would still like to figure out why this doess't work so that I could learn something from it.