Add peers-aware PUSH/PULL throughput rows

[rgbkrk]

Summary

This creates a fresh benchmark-only PR inspired by the old #263 cloud-agent branch.

Changes:

• add peers-aware throughput/push_pull/{transport}/peers={peers}/{size} rows for zmq.rs and libzmq
• add a focused pushpull-omq perf-suite profile that matches the OMQ PUSH/PULL peer and message-size grid
• include the key peers=8,size=8192 PUSH/PULL rows in the standard profile
• teach scripts/run_perf_suite.py to expand {peers} filters and avoid artifact collisions when a profile has multiple entries for the same bench target
• add copied/refreshed unit coverage for perf-suite filter expansion and artifact naming
• keep the new libzmq PUSH/PULL benchmark workers bounded: receive timeouts prevent hangs and worker threads are joined at teardown

Notes

This is measurement coverage only. It does not change socket behavior or claim a new performance result.

Validation

rustfmt --edition 2021 --check benches/throughput.rs
python3 -m json.tool perf-suite.json >/dev/null
python3 -m unittest scripts.test_run_perf_suite
git diff --check
cargo bench --bench throughput --no-run
cargo bench --bench throughput --no-run --no-default-features --features async-std-runtime,all-transport
cargo bench --bench throughput --no-run --no-default-features --features async-dispatcher-runtime,all-transport,async-dispatcher-macros
python3 scripts/run_perf_suite.py --dry-run --profile pushpull-omq --impl zmqrs,libzmq --transport tcp --run-id dry-run-pushpull-omq --repo-root .