3nb-spr iavf: correlated anomalies in ipsecswasync tests

[vrpolakatcisco]

Most tests are unaffected, or have too high noise. But one pair of tests show anti-correlated anomalies in trending [0]. One has regression when the other has progression and vice versa.
I suspect this is some quirk of compiler, and impact is small, but bisecting could still be hopeful, as #4099 had similar anomalies but is harder to bisect.

[0] 2c aes256gcm and 3c aes128cbc-hmac512sha

[vrpolakatcisco]

Bisect found and verify confirmed this [1] is the cause of the late-february performance change. As some tests show progression and looking at the code, this is not unexpected, so probably fine. Still searching for the second apparent regression.

[1] 42366: ipsec: combine huge and normal anti-replay-window handling | https://gerrit.fd.io/r/c/vpp/+/42366

[vrpolakatcisco]

After few unstable runs, one bisect arrived at [2] as the second anomaly. I think it is right, I have seen that as a probable cause for few anomalies, too small a regression to investigate further.

[2] 42537: vlib: add new node type - SCHED nodes | https://gerrit.fd.io/r/c/vpp/+/42537