fix(google_benchmark): dont start and stop valgrind twice

[not-matthias]

No description provided.

[codspeed-hq]

CodSpeed Performance Report

Merging this PR will not alter performance

_{Comparing cod-1963-codspeed-cpp-starts-valgrind-twice (56e942a) with main (5ef9067)}

Summary

✅ 192 untouched benchmarks

[GuillaumeLagrange]

This is handled by the iterator impl right ?

Could you add a tiny bit more context as to what this fixes and why we're fixing this ?

[not-matthias]

This is handled by the iterator impl right ?

yes, we're already starting/stopping in the State which is used in the benchmark loop.

Could you add a tiny bit more context as to what this fixes and why we're fixing this ?

The main issue is that in simulation mode, we were calling measurement_start/stop twice. For now, this didn't cause any issues because we're clearing the events when starting, and nothing happens when stopping.

I believe we should remove it just to prevent errors later on, in case we're extending the valgrind implementation or when reusing it for memory profiling for example.