- It was only available in the ia32 backend and broken there. As the
  feature is rarely used, I remove it for now instead of going through
  the trouble of fixing it.

The previous approach put the irg link into the block attributes, so
while saving memory an extra indirection was necessary. Some testing
with valgrind massif+callgrind revealed that a direct link increses peak
memory usage by ~1% while also increasing runtime by ~1%. More
importantly the code gets a bit simpler and some strange corner cases can be
removed with the direct link.

This fixes opt/fehler266.c.

This makes our usage of PhiMs to represent the observable behaviour of
non-terminating loops more obvious.
Having an explicit loop Phi constructor also fixes problems with Phi
getting optimized before the keep edge was added.

Settle on 1 flag type and no different I_FLAG and E_FLAG

The reassoc running flag wasn't necessary in new reassoc versions
anyway.

A self-loop before the first real value was not recognized before and
could lead to an invalid kept-PhiM removal.

For details see endless loops section on the homepage.

- Mux((a & 1) != 0, 0, b) => -a & b
- Mux((a & 1) == 0, 0, b) => (a - 1) & b