Incorrect branch coverage when running two async iterators

[sfstpala]

Describe the bug
When I run two (or more) async iterators in async for loops, the second loop will have incorrect branch coverage.

To Reproduce
Here's an example:

import asyncio

class AsyncIter:

    def __init__(self, items):
        self.items = items

    async def __aiter__(self):
        for i in self.items:
            yield i

async def test():
    async for i in AsyncIter([1, 2, 3]):
        pass
    async for i in AsyncIter([1, 2, 3]):
        pass

asyncio.run(test())

And the result:

I'm running Coverage 5.5 on Python 3.9 (debug sys output here)

Expected behavior
Same as the non-async version:

[sfstpala]

I get the same issue when I just put the async for in another loop like this:

async def test():
    for i in [1]:
        async for i in AsyncIter([1, 2, 3]):
            pass

[sfstpala]

I also noticed that this works fine in Python 3.8, but not in Python 3.9 and 3.10.

[nedbat]

See this comment on #1158.

[nedbat]

I think this is a combination of two CPython bugs:
async-for loops are traced incorrectly in Python 3.10 and Python 3.9 traces async for/else incorrectly.