> So, text()[not(following-sibling::node())] may match the last text
> element. You could use normalize-spaces but that will also
> remove leading spaces.

prefix the text() with a char and substring from the 2nd position.

   substring(normalize-space(concat('X', text())), 2)

That's an old trick I've seen here alot.  It will save at least 1 leading space.