This adds two new APIs:

```rs
turbo_tasks::emit(some_trait_vc)
```

emits some kind of value to the current task. It will bubble up the call graph until processed via:

```rs
let some_vc = some_function();
let all_emitted_vcs: Vec<SomeTraitVc> = some_vc.peek_collectibles::<SomeTraitVc>().await?;
// or
let all_emitted_vcs: Vec<SomeTraitVc> = some_vc.take_collectibles::<SomeTraitVc>().await?;
```

The `peek` variant will return the emitted values, but they will continue to bubble up the call graph.
The `take` variant will stop bubbling and parent tasks will not see these values. You may re-emit them via `emit()` again.
Peeking is a bit more efficient.

Emitting is pretty cheap. Peeking and taking is counted as read operation and the task will invalidate when new values are emitted or existing ones are removed.

---

This also improves the error handling for node-file-trace:

![image](https://user-images.githubusercontent.com/1365881/182683991-84c6dfe1-effc-4433-9f3c-c12e51828f47.png)

but there is still a lot to do to improve that...