The main time a dominator search in the cache fails to produce an 
accurate range is when we have a "complicated" join node.

ie

bb4:
    if (a > 200) goto bb5 ; else goto bb9
bb5
    if (b > 400) goto bb6 ; else goto bb7
bb6
    foo (b)
bb7
    onward()
    goto bb200;
bb9:
   if (b > 200) goto bb6; else goto bb200

in this case bb6  is a complicated join node by ranger standards. It has 
2 incoming edges which carry range information for b:
    9->6 has [201, +INF] and
    5->6 has[401, +INF]
The immediate dominator of bb6 is bb4.  When asking bb6 for a range from 
its dominator, there is no sign of any outgoing ranges for b, and 
range_from_dom would simply return VARYING.

This patch tweaks range_from_dom()  so that we calculate the range for 
any block in which either
   1) The dominator has an outgoing range    (this was the old behaviour 
only)
   2) The current block sees an outgoing range on an incoming edge.
The addition of 2) allows us to capture the above case with minimal cost 
which gets use very very close to the results from the old style full 
cache value propagation.

Bootstrapped on x86_64-pc-linux-gnu with no regressions.  pushed.

Andrew