This PR partially fixes #2744

This PR makes two changes:

1. Call inspect.stack(0) instead of inspect.stack()
2. Only call inspect.stack() when getattr(task, "__pw_stack__") doesn't find a stack in the task

inspect.stack() by default requests context for the first frame of the callstack, which is a very expensive operation. We don't use the returned context information, so we can safely use inspect.stack(context=0).

In my testing, this improved the speed of evaluate() by about 2.5x:

Test:

async def test_speed_test(page: Page) -> None:
    start_time = time.time()

    for i in range(0, 1000):
        await page.evaluate('() => 1 + 1')

    print(time.time() - start_time)

Before this PR: 10s
After this PR: 4s

The second change is that we originally were calling getattr(task, "__pw_stack__", inspect.stack()), which retrieved the entire callstack and then threw it away if getattr(task, "__pw_stack__") succeeded. I wasn't able to figure out how to get getattr(task, "__pw_stack__") to return a value, so I'm not sure how much this change sped things up, but it seems like a sensible change to make.

I've added async and sync tests.

I have a more detailed fix, but I figured I'd first try to get this PR landed, since it involves small changes for a pretty large speedup :)