New Exchanger Implementation

[jdantonio]

Brand new implementation based on OpenJDK 8

[jdantonio]

I believe this new implementation is functionally correct. Both the original condition variable implementation and the later MVar implementation had known bugs, which is why we moved Exchanger into Edge. So I don't think there is any harm in merging it, even if we discover bugs later.

@mighe @chrisseaton @pitr-ch I know you are all very busy, but if any of you have a chance to review this PR I would appreciate it.

[chrisseaton]

The crucial method for anyone else reviewing this: https://github.com/ruby-concurrency/concurrent-ruby/pull/396/files#diff-8a5aa6ad7f52e71c4ce9a64573329694R163

I don't understand the high-level algorithm here so it's hard to verify that it's a correct algorithm, or that it's correctly implemented. Can you describe the algorithm in a comment, or reference the code where it is from?

[jdantonio]

@chrisseaton Done. The algorithm is described (in detail) in the source. Once I wrote it all out I discovered a bug so that was a bonus.

[mighe]

I've read carefully both the algorithm and the code, I've also run the code "by hand" looking for race conditions.
It seems ok, I could not find any issue, but I'd like to review again everything on Monday to be sure that it is really ok :)

[jdantonio]

This PR has been rebased against the latest master (after thread_safe was merged). I would like to include this PR in the 1.0.0.pre1 release but that is not essential. I'll leave it open for a couple of days to get more review/feedback. We'll release 1.0.0.pre1 next week with or without this PR.