Investigate too low perf gain from NonValidated headers API

[alnikola]

We switched YARP header processing to the new HttpHeaders.NonValidated API and measured how RPS changes. RPS grew up but only by 2% whereas in the previous experiments with a reflection based approach to non-validated headers enumeration the 7% RPS increase was observed. We need to investigate why there is such big difference between the expected increase (7%) and the actual increase (2%).