Rework task::try for new task_builder interface.

[bblum]

Maybe users want task::task().supervised().try. The current toplevel one should stay the same - i.e., equivalent to a potential new task::task().unlinked().try.

[brson]

Maybe it does make sense for task::try to to supervise the child. There might be some way for the parent to be killed but the unlinked child to keep running under the current scheme, which sounds potentially errorsome.

[bblum]

Yeah, I think that makes more sense as a default behaviour. I just wrote f57760c, for which try would work either way. I guess I'd need to audit all other callsites.