Almost removing oldcomm

[brson]

r? @graydon

This branch removes most uses of oldcomm. It isn't quite ready but I've been at it for a while so I'm posting it for review. Removing the last few usages of oldcomm from std is proving to be time consuming because there's a lot of tricky unsafe code involved. This also won't be mergable until I figure out #4478.

The general strategy:

• Fix New tasks should always spawn into the default (general) scheduler unless otherwise requested #3760 so task spawning has sane behavior. These changes are going to add some more 'service' tasks that spawn lazily and they should not spawn onto random schedulers.
• Add a general at_exit facility where the runtime will execute a closure just before shutdown.
• Add a simple, unsafe 'global data' API. It is simply a way to access a global hash table under a mutex. Uses at_exit for cleanup.
• Add a new weak task module that uses pipes, at_exit and global data
• Convert users of the old weak task code and chan_from_global_ptr to the new interfaces
• Remove the old weak task and chan_from_global_ptr functions
• Remove oldcomm

Organizationally, this dumps all the new modules under core::private, which I recognize is not right. I intend to reorganize the runtime code as outlined in #2240.

[brson]

Oh, this also sneaks in my try/finally library, described here, though it's hidden under the private mod.

[graydon]

Fantastic. r+ with 100 +s