rt: unnecessary calls to unpark I/O driver

[carllerche]

I'm filing this as a runtime issue though the symptom is invocations to the I/O driver unpark method.

I have identified unnecessary calls to the time driver's unpark() method, which translates to calls to the I/O driver waker. The calls are unnecessary due to the I/O driver not currently being parked. Each call driver unpark call is relatively expensive.

Each time::sleep() will potentially unpark the time driver. This is so that if the driver is parked, it can observe the new timeout. This unpark always happens even when using the current-thread runtime. The unpark is unnecessary because the I/O driver is not currently parked.