Align benchmarks with hyperlight-core and the benchmarks repo #44

Benchmarks already exist. hluk bench runs cold, cold-snap, warm-restore, warm-stateful, and parallel modes; benchmarks/limits.json sets per-runtime, per-OS ceilings that gate CI; and results are charted per host OS via github-action-benchmark on gh-pages. The remaining work is deeper alignment with how hyperlight-core measures performance and with the shared benchmarks repo, so numbers are comparable across projects (methodology, metric names, and reporting). hyperlight-dev/hyperlight-bench#30 adds the hluk runtimes there. arm64 hosts are not benchmarked yet.