2n-grc: rarely, Linux is too slow to re-detect interfaces after VPP kill #4113

The symptom visible in console logs is Failed to set flow control on {pf_eth}!, but log.html [0] shows the true issue is a command like basename /sys/bus/pci/devices/0000:01:00.1/net/* returning an empty string instead of a valid Linux interface name the subsequent commands require.

I am not sure if other testbeds are also affected, as the occurrence is very rare.
I believe this issue was already causing the symptom previously reported as #4074, but back then I thought the old VPP instance was still hogging the interface. This time the log proves it was more than 1.4 seconds between VPP confirmed killed and interface still not detected, so I guess this is some infra issue or driver issue, in any case too nondeterministic to verify properly.

[0] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-iterative-2510-2n-grc/36/log.html.gz#s1-s1-s1-s1-s2-t2-k2-k4-k1-k1-k1-k1-k1