Invalidate the block on call completion

[halorgium]

When calling a method on an Actor from a normal Thread, if you invoke a block after the Call has completed, you will encounter a deadlock.

This is due to the fact the sending Thread is no longer waiting for the block invocations.
gist

If you are calling from an Actor, this behaviour works.
gist

This change invalidates the block for both scenarios.
We can allow block invocations in the actor case by only invalidating if the sender is a normal Thread.

• Specs
• Remove extra FIXMEs
• Consider async/future + blocks with respect to invalidation

[coveralls]

Coverage increased (+0%) when pulling b3422fa on invalidate-block-on-call-complete into 963245c on master.

[halorgium]

This should probably be backported to 0.14 too.
The issue was originally reported there and will cause confusion.

[tarcieri]

@halorgium what's the story with this?

[halorgium]

@tarcieri basically, the current patch is overly (?) aggressive in disallowing use of the block.

I think we should allow use of a block after the initial method call has been completed.
This does mean that if the actor which originally invoked the method dies, the call to the block could potentially hang.
Given that we could check the mailbox is still alive before sending the BlockCall as we do for SyncCall.

[digitalextremist]

@niamster do you feel like this could be part of the same thinking behind resolving #463?

[tarcieri]

This is woefully out of date. Closing.