perf: speed up Python stack capture

[Skn0tt]

2.0–2.6× faster Playwright calls under pytest; 1.2–1.7× faster standalone.

This replaces inspect.stack(0) with direct f_back traversal. It still captures the full stack, but avoids eagerly constructing FrameInfo objects and inspecting source context on every API call.

Paired page.evaluate("1") benchmarks on Python 3.12:

	Sync	Async
Standalone	1.16×	1.70×
pytest	2.01×	2.56×

The difference is larger under pytest because its runner stack is deeper.

Re #2836.