bug#52491: 28.0.90; Regression in window deletion with minibuffer

[Juri Linkov <juri@linkov.net>]

>  M-x	    ;; execute-extended-command
>  <tab>	    ;; minibuffer-complete
>  <kp-prior> ;; switch-to-completions
>  q	    ;; quit-window
>
> Point moves to the *scratch* buffer.  This is a regression.
> In emacs-27 it moved to the minibuffer.
>
> The problem is that delete-window now uses get-mru-window
> to get the window to select after deleting the completions window.
> But get-mru-window ignores the minibuffer window.

There is another problem:

 M-x	    ;; execute-extended-command
 <kp-prior> ;; switch-to-completions
 C--	    ;; negative-argument
 C-x o	    ;; other-window
 C-x o	    ;; other-window
 q	    ;; quit-window

After switching to the *scratch* buffer with ‘C-- C-x o’, and back to
the *Completions* buffer with ‘C-x o’, then ‘q’ (quit-window) selects
the *scratch* window because it's the most-recently-used window.
But in 27.2 it selected the minibuffer window.