Projs are always in the same block as their predecessor.

sched_get_time_step() checks the same.

make it more robust and remove unnecessary req check

This also corrects dumping schedule edges.

- split edges_notify_edge into add/del/modify and call the correct
  function where we know which is needed
- get rid of some unnecessary flags so the ir_edge_t struct is smaller

- save ir_node* instead of irn_idx in the map (saves irn_idx_map
  lookups)
- in liveness_transfer check register class on the inputs (which is
  cheaper) and only then check the operand

We have simpler code without it, and a heuristic based on
get_irg_last_idx() seems to work as well.

We get a clean SIGSEGV anyway, if they're not fulfilled

So all files in libFirm do this consistently

Profiling uses a magic function named mcount.
When profiling libFirm, these two symbols with the same name lead to strange problems.

rbitset: Handle the edge case that the start position equals the end position in rbitset_next_max().

This fixes rbitset_foreach().

It only measures initializing a single struct.