Expose runtime apis

Expose more runtime internals so that users can implement custom synchronization APIs.

As an example, take [atomic-wait](https://crates.io/crates/atomic-wait). This crate provides a futuex-like api for all modern operating systems, and has a few million recent downloads on crates.io. Unfortunately, testing code that uses this crate is difficult with shuttle. While it is possible to replicate something like the functionality of a futex using a mutex and condvar, I don't think the runtime ordering effects are exactly identical. I'd prefer to be able to implement the futex functionality myself with `thread::switch` and similar, so I can be confident that my tests are modeling the actual semantics of my synchronization primitives.

The issue names the `thread::switch` entry point and the `atomic-wait` crate, but no files or tests. Start by reading the runtime API around `thread::switch` and determine which internals a custom futex-like synchronization API would need. Done means users can implement that API while preserving the runtime ordering semantics they need to test.

Written by the indexing model from the issue text.