1) While marking the backing hash table of a WeakMap we also need to
   record the slot because it might be on an evacuation candidate.
2) With incremental marking one backing hash table might be marked more
   than once because the WeakMap might have gone through a white to gray
   transition.
3) The corner case when the allocation of the backing hash table itself
   causes a GC, leads to a WeakMap with an undefined table field, so we
   still need to handle this case correctly.

R=vegorov@chromium.org
TEST=mjsunit/harmony/proxies-example-membrane --stress-compaction

Review URL: https://chromiumcodereview.appspot.com/9985010

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@11385 ce2b1a6d-e550-0410-aec6-3dcde31c8c00