time: polling sleep future after complete does not panic in release mode.

[carllerche]

From: https://fasterthanli.me/articles/pin-and-suffering (search for debug_assert!).

use futures::Future;
use std::{mem::swap, pin::Pin, task::Poll, time::Duration};
use tokio::{macros::support::poll_fn, time::sleep};

#[tokio::main]
async fn main() {
    let mut sleep1 = sleep(Duration::from_secs(1));
    let mut sleep2 = sleep(Duration::from_secs(1));

    {
        // let's use `sleep1` pinned exactly _once_
        let mut sleep1 = unsafe { Pin::new_unchecked(&mut sleep1) };

        // this creates a future whose poll method is the closure argument
        poll_fn(|cx| {
            // we poll `sleep1` once, throwing away the result...
            let _ = sleep1.as_mut().poll(cx);

            // ...and resolve immediately
            Poll::Ready(())
        })
        .await;
    }

    // then, let's use `sleep1` unpinned:
    swap(&mut sleep1, &mut sleep2);
    // by this point, `sleep1` has switched places with `sleep2`

    // finally, let's await both sleep1 and sleep2
    sleep1.await;
    sleep2.await;
}

Results in:

thread 'tokio-runtime-worker' panicked at 'assertion failed: cur_state < STATE_MIN_VALUE', /home/amos/.cargo/registry/src/github.com-1ecc6299db9ec823/tokio-1.4.0/src/time/driver/entry.rs:174:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

But the panic is from a debug assert: debug_assert!(cur_state < STATE_MIN_VALUE);, so in --release mode, nothing happens.

Additionally, the panic message is poor. We should:

a) panic in release mode as well
b) have a better error message describing the problem.

[Darksonn]

Note that the code that was posted has undefined behavior. Of course, it's still nice to panic with a nice message if possible.