[master] RFR: Parallel: fix size_given_klass

[Stefan Karlsson]

The Parallel scavenge threads compete to copy and forward objects, and the winning object is the one that managed to update the the forwarding pointer. For Lilliput, this code needs to be extra carefully written because installing the forwarding pointer destroys the klass pointer. The threads need to read the object header once, and then use that information when figuring out the size of the object. The code has been changed to do this, but there's a place in an assert were the code tries to reread the klass pointer, and then crash because it has been racingly been overwritten by another thread.

I had an idea that we could fix this by simplifying the assert with #149, but that has a problem with self-forwarded objects. There are ways to fix the problem listed in that PR, but for now the suggestion is to just make sure that the asserting code doesn't reread the klass pointer.

-------------

Commit messages:
 - Parallel: fix size_given_klass

Changes: https://git.openjdk.org/lilliput/pull/163/files
  Webrev: https://webrevs.openjdk.org/?repo=lilliput&pr=163&range=00
  Stats: 4 lines in 3 files changed: 0 ins; 0 del; 4 mod
  Patch: https://git.openjdk.org/lilliput/pull/163.diff
  Fetch: git fetch https://git.openjdk.org/lilliput.git pull/163/head:pull/163

PR: https://git.openjdk.org/lilliput/pull/163