- Previously it was part of load-store optimisations, the rewrite now
works as a localopt.
- The new version works for address arithmetic and not for Sels anymore
  (while the old code only worked for Sels)
- The new version can handle some forms of initializer spanning loads
  like loading a string like "bla" with an mode_Iu to get all 3 chars
  and the zero in one load.

also move copy_const_value from typerep/entity to irgmod.

This is necessary because the walker does not guarantee that all Projs
of the second Tuple are visited after we optimized the first Tuple.

No need to do it in backend preparation phase, if some backends destroy
the property again before transformation.

This also fixes a bug where irout computation would claim that it
removed all tuples, but missed tuples just held by a keepalive edge.

This fixes opt/ifconv5.c and 400.perlbench.

This fix is necessary to avoid an assert in the new
itfield/compound_member code.

We do not model them as a type anymore, but simply annotate
compound_member entities with a bit offset and bit size.

In the case of a Rol/Ror pattern the load value is used two times (by
the Shl+Shr), destAM matcher always matched for one user only.

Also added a new matcher for the ia32 backend matching Shl(x,1) to
Lea(x,x).

This mainly adds comments explaining the situation and avoids an extra
graph walk.

This works around a problem with modeling control flow for calls to
setjmp and similar functions.

This reverts commit 3d1d2c43.

Creating a Bad node does not imply that it has users or is reachable
within an irg. While clearing the property is certainly correct as a
conservative approximation, it should not be necessary.