Determine panic and abort policy

[NthTensor]

We currently don't handle panics or aborts in any-way. We should set out and implement a policy for responding to panics. It's likely we will just want to copy how rayon does this.

[NthTensor]

This should be largely resolved by #27. As of that PR, join will catch and propagate errors back to the calling thread, and everything else just aborts. As follow up we should also try to have scopes propagate panics, but that can be it's own issue.