Recursive logging from #4799 performance

[aabmass]

#4799 is not yet released but it works by walking the stack on each on_emit() call which is pretty hot code.

I think using a contextvar (or OTel context) would be significantly faster since loading a traceback is slow and then we have to walk the stack. Rough estimate of perf difference:

$ python -m timeit -s 'import traceback' 'sum(
                item.name == "on_emit"
                and (
                    item.filename.endswith("export/__init__.py")
                    or item.filename.endswith(
                        r"export\__init__.py"
                    )  # backward slash on windows..
                )
                for item in traceback.extract_stack()
            )'

10000 loops, best of 5: 22.8 usec per loop

$ python -m timeit -s 'import contextvars; c = contextvars.ContextVar("c", default=False)' 'token = c.set("True"); bool(c.get()); c.reset(token)'
2000000 loops, best of 5: 157 nsec per loop