asyncio extension

[bdunahu]

Currently, it is impossible for Scalene to report over 100% for a program; native code which executes in parallel should still be associated with a single line of python, and even if this were not the case (this applies to the multiprocessing library too), the samples can only be accounted for if they are present in the new_frames variable during each sample, each of which are currently treated as separate threads and assigned a normalized time based on how many there are.

This means the reporting problem is a new issue: it would be wrong to add the idle frames to new_frames, because that would imply they block the python interpreter from doing anything else while they're waiting.

When an asynchronous task is suspended, we instead assume it is waiting for the entire sampling interval. The solution this PR implements for the reporting problem is to treat idle tasks as if they run sequentially after non-waiting code, one after the other (i.e., nothing truly happens in parallel). This means the total CPU time passed is adjusted to match every sample.

Because asynchronous tasks run regardless of what the GIL is doing, the results are usually biased towards asynchronous code. It still leads to the behavior most users would likely expect.

Current state:

• idle-task frame collection logic is implemented in scalene_asyncio
• it is possible to prevent time from being assigned to the asyncio event loop by filtering out frames which belong to a thread that is running an event loop but has no current task. However, I do not do this, because throwing out frames complicates the above approach to reporting results.
• the passing of Scalene.should_trace is hacky. This function is also passed to scalene_utility.add_stack, is it possible for the implementation be moved to the utility file?

[jaltmayerpizzorno]

Preliminary review while I wait for an answer to a question I posed on Slack.