Drop unneeded profiling events for timing accuracy

[linamy85]

We have observed large data size and topology microbenchmark may yield large profiling trace exceeding the limit and hence lead to inaccurate timing.

[chishuen]

Are you able to test if this will break other microbenchmarks (e.g. gemm, etc.)?

[linamy85]

Are you able to test if this will break other microbenchmarks (e.g. gemm, etc.)?

Yes, this was verified on Ironwood using other microbenchmarks. The numeric differences compared to the current main branch are considered within variance.