In our quest for performance, the performance of sys.settracing based tools has probably gotten worse.

1. How do we measure this?
2. How do fix this?

We will initially use coverage.py as proxy for all sys.settracing based tools when measuring  performance.

The fix is probably to use quickening to insert a minimum set of instrumentation instructions required for tracing/profiling.
The existence of `f_trace_opcode` is a bit of a problem however, as we will have to instrument *every* instruction.


Ideally, sys.settracing based tools should be faster on 3.11 than 3.10, but at the least we should provide a simple alternative to sys.settracing that is faster.