The jump optimization maybe run Turbofan pipeline twice for each TF/CS builtins,
and relies on the fact that the number of j/jmp instruction generated is always
the same.
The behavior of {AddMatcher::SwapInputs} should be aware the two times code
generation, and prevents the flipping of child nodes.

For example:

1: Int32Add(2, 3)   --- We shouldn't swap the input #2 and #3 in this situation
2: Int32Sub(4, 5)
3: Int32Add(6, 7)
4: ...
5: ...
6: ...
7: ...

R=danno@chromium.org

Bug: v8:7839
Change-Id: Ia97de3ab28294e595ac27b5898c099c0d782e9f9
Reviewed-on: https://chromium-review.googlesource.com/1098678Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Kanghua Yu <kanghua.yu@intel.com>
Cr-Commit-Position: refs/heads/master@{#53705}