fix(benchmark): a pcap engine's loader OSError kills the whole environment on macOS and Windows #2129

On macOS and Windows runners, the benchmark's 3.10 and 3.11 pcap_ct environments die before measuring anything. The cause is that examples/benchmark/benchmark.py's _unavailable doesn't recognise the loader errors pcap-ct raises there. Seen in the #2128 runner-comparison run, 40047570595.

The error per runner:

• macOS: OSError: dlopen(libc.so.6 …). pcap-ct's libpcap loader looks for a Linux libc.
• Windows: OSError: Cannot find npcap/wpcap.dll, because Chocolatey's Wireshark doesn't install Npcap.

Why it kills the environment:

• _unavailable turns an OSError into an explained "not measured" reason only when its message contains libpcap.
• These two messages don't, so preflight("pypcap") in the pcap_ct environment re-raises and the whole environment exits.
• On 3.12+, the not-installable note answers the pypcap check before any import, so those environments survive.

Effects:

• The 3.10/3.11 columns rest on half their samples.
• The reason recorded for pcap_ct is "package is not installed" rather than the real loader error.

Fix: treat an OSError raised while importing or opening a pcap engine as unavailable, with its message as the reason, whatever the wording. The engine then becomes an explained row, and the environment carries on with the other engines. Add a test with each of the two messages.