[RFC] handle runtime shutdowns while tasks are being spawned

[sunshowers]

If a runtime has been marked as shutting down before spawn_blocking is
called, or before the blocking task is first scheduled, the JoinHandle will
return a cancellation error. Currently, this panics, which tends to manifest as
test failures.

Instead, add a RunError wrapper that captures the cancellation situation.
Callers might choose to bubble this error up or to squelch it -- unfortunately,
we can't make this decision.

I haven't written tests yet but I should be able to if this looks reasonble.

The API surface here is really unfortunate, but I can't see a better way to do
this than to propagate errors up the stack.

[sunshowers]

oh my god what's wrong with you spr.