The `next_retry_time` gets populated when a request receives an error
timeout or any other error, after thatn next_retry would check all
requests in the queue returns the smallest one, which then gets used to
move the main loop by creating a Delay
```
futures_timer::Delay::new(instant.saturating_duration_since(Instant::now())).await,
```

However when we retry a task for the first time we still keep it in the
queue an mark it as in flight so its next_retry_time would be the oldest
and it would be small than `now`, so the Delay will always triggers, so
that would make the main loop essentially busy wait untill we received a
response for the retry request.

Fix this by excluding the tasks that are already in-flight.

---------

Signed-off-by: Alexandru Gheorghe <[email protected]>
Co-authored-by: Andrei Sandu <[email protected]>