bug#43389: 28.0.50; Emacs memory leaks using hard disk all time

[Trevor Bentley]

> > > Does thread-last-error return something non-nil?  
> >  Nope, nil in all instance, including the one in a weird state. 
>
> Then it's unlikely that a thread died unnatural death. 

No, sure doesn't seem like it.  Just hit it in an instance with 
more printfs, and it looks like it leaps right out of some 
sub-call of process_pending_signals(), continuing to run elsewhere 
without finishing garbage_collect().  To me, that means exactly 
one thing: longjmp.

If something manages to longjmp out of garbage_collect() at that 
point, it leaves with consing_until_gc set to HI_THRESHOLD.  This 
must explain why automatic GC stops running for hours or days, but 
manual GCs still work.

I tried setting a breakpoint in longjmp, but it's called 3 times 
for every keypress!  That's inconvenient.  Running one single 
instance now with a conditional breakpoint on longjmp: it will 
break if longjmp is called while it's in unblock_input().

-Trevor