(Note: I originally mentioned this in bug#51993, but it's only 
somewhat-related to that bug. To make things easier to follow, I pulled 
this part out into a separate bug.)

To see this in action:

   $ emacs -Q -f server-start
   $ emacsclient -c foo.txt

   ;; From the first (non-client) frame:
   C-x 5 0  ;; delete-frame

   ;; From the second (client) frame:
   M-x kill-emacs
   ;; Emacs prompts:
   Error (error Attempt to delete the sole visible or iconified frame); 
continue? (y or n)

Pressing "y" will properly kill Emacs, but there's no real use for the 
warning in this case. It happens because 'server-force-stop' 
(indirectly) tries to delete the last (client) frame before killing 
Emacs, meaning that Emacs would have zero frames for a bit.

Since 'server-force-stop's docstring says, "This function is meant to be 
called from `kill-emacs-hook'," it should be safe to have 
'server-force-stop' avoid deleting any frames: they'll just get deleted 
when Emacs is actually killed.

Attached is a patch to do this. Note that I named the new argument 
"noframe" because that matches the existing code in server.el (see 
'server-delete-client'). It's a bit of a misnomer though, and maybe 
"keep-frames" would be better...