split_edge improvement

[Jeffrey A Law]

  In message < 19991117232355.A29164@cygnus.com >you write:
  > Jan has had a similar patch floating about for a little bit.
  > One of the things I didn't like about it (which I should have
  > told him -- shame on me) is that it assumed the very previous
  > insn was the loop note.  No other intervening notes allowed.
  > 
  > If either of you can show that there can't (or shouldn't be)
  > any such intervening notes, I would be much happier.
stmt.c generates the note immediately followed by the loop start
label.  The only way we can get junk between them is via one of the
optimizers.

jeff