Reduce Profiler overhead #9225

The goal is to reduce the performance impact of profiling through two complementary approaches:

1. Optimize the hot path: Reduce the overhead of collecting timing samples by making the profiling code faster and minimizing work performed during each sample. Some of these changes are also expected to slightly improve the performance of non-profiled code.
2. Add asynchronous sampling (default mode): Introduce a new profiling mode that collects timing samples from a separate thread, reducing the instrumentation overhead on the profiled attachment. This should become the default mode.

The existing synchronous sampling mode should remain available as an alternative.