More robust treatment of uncaught panics for child threads

[aturon]

Currently, an uncaught panic on a child thread will result in output to that thread's panic output handle (which is currently overridable).

This may not be a desirable default, and we probably want the behavior to be more thoroughly customizable.

See this previous RFC for some discussion on the topic.

[aturon]

Note: a change to the default would likely be a breaking change.

[blaenk]

Yup.