Discuss/determine laws for ParFuture & ParIVar

[rrnewton]

UPDATE: This is important enough that I moved it to a Wiki page.

We could still use this issue as a DISCUSSION THREAD for that wiki page.

[rrnewton]

One point of tension here is that the idempotence-requiring schedulers can ONLY implement the idempotent APIs.

If IdempotentParIVar is a subclass of ParIVar that's actually backwards. The traditional LVish idempotency-assuming scheduler could implement the former interface but not the latter. That is, if it implemented the latter random multiple-put exceptions could happen at any time.

(Of course, a fully correct IdempotentParIVar requires VerifiedEq, which we don't have yet.)

We should probably start using VerifiedEq -- just as an alias at first -- to keep track of where we need it.