When data after close is on, you can mess up the reassm queue

[rrs]

Summary

If you are using the data-after-close feature it is possible
you will borf the reassembly queue. This happens due to
the fact that once we set the flag we still push zero len
segments into the reasm without this fix.

Test Plan

Tested at NF and will also test with head.

[kbowling]

Jason Eggleston and I have looked at this and it makes sense to us