reload vs adds with clobbers

[DJ Delorie]

> Yes.  In fact, I thought that was already done for xstormy16, but I
> probably missed a case...

I might have messed it up, too.  I'm working on a newly reported bug,
and the fix for that broke the built-in-setjump test, leading me to
this point.  FYI, the new bug is a case where the addhi3 is created
with the expander, but has the (scratch:HI) in it - after reload has
begun, so it doesn't match the pattern's constraints.