Use a .start() method?

[Dreamsorcerer]

Related to #69.

I'm wondering if we should move the creation of the wait_failed task to a .start() method or similar, so that it is not a mistake to instantiate the Scheduler outside of a running loop.

So valid code could look like:

scheduler = Scheduler()

async def main():
    scheduler.start()
    await scheduler.spawn(coro())

I'd like to hear some thoughts from @asvetlov, so leaving this here.

[asvetlov]

Well, why not?

[asvetlov]

.start() shouldn't be a part of public API IMHO, we can do lazy initialization with check for the same loop in async functions, like asyncio does for locks.