bug#56393: Actually fix the long lines display bug

[Gregory Heytings]

> > (Note that I did not actually try the above recipe, it's just meant to 
> > show that there will always be corner cases that won't be handled, even 
> > with a dynamic detection of long lines.)
>
> I think arbitrarily limiting C-p in buffers that don't have long lines 
> will be deemed a regression.

I bet that we won't have a single bug report about this.