- Compound type selection is performed with Member
- Array indexing is performed with Sel
- Instead of the simpleSel constructor functions you can use the normal Member
  constructors now
- The concept of array element entities has been removed, Sel nodes just
  contain a pointer to the array type now.

- is_address_taken() according to its own logic was only valid when
  called with a Sel node
- Tuples where skipped although irg_properties ensured no tuples in the
  graph anyway.
- Call and default case could be merged.
- Id case would have failed if taken (is_address_taken called with Id
  instead of Sel)

This also disables the memcombine optimisation if unaligned stores are
not supported by the backend.
We may add some tests in the future to reenable this in cases where we
can prove that the combined store is aligned.

We should continue here as this is probably just an unsigned loop going
backwards (firm normalizes -1 to +0xFFFFFFFF on unsigned modes). How to
fix this properly is left as an exercise for the next developer :)

Rework load memory chain walking:
- We keep the pointer as a base + offset representation all the time
- Reuse predict_load code from localopt for more accurate replacement of
  Loads from constant entities
- Support some more cases where load-after-load and load-after-store did
  not use the same mode but where we can easily get the value with some
  arithmetic.

This fixes opt/loadstore6.c.

This fixes gcc/pr36988.c

- Does not invalidate graph properties anymore if nothing changed
- Only produce Bad/Jmp Tuple predecessors for possible throwing
  Load/Store nodes
- Improve check_load_store() logic slightly and rename to
  is_conv_bitcast()

The initial block is not immature after new_ir_graph() anymore for a long time.
The end block is matured automatically by irg_finalize_cons().

This is the only part, which is used of the entry_edge.

It is not sensible to count Projs.
They only represent an aspect of another node.