perf: add reproducible private and group messaging benchmarks

[yuwnloyblog]

Summary

• add a configurable jimbench CLI using the production WebSocket protocol
• cover private-ring and group fan-out workloads with separate connection, ACK, and delivery metrics
• report throughput plus P50/P95/P99 latency, errors, workload settings, commit, OS, CPU, memory, database, and deployment labels
• add a guarded one-command Docker Compose runner and machine-readable JSON results
• document methodology, limitations, reproduction, and publication requirements
• add benchmark tests to CI

Verification

• go test ./simulator/benchmark ./cmd/jimbench
• go test -race ./simulator/benchmark
• go vet -composites=false ./simulator/benchmark ./cmd/jimbench ./simulator/wsclients
• CGO_ENABLED=0 go build ./cmd/jimbench
• bash -n scripts/run-benchmark.sh
• docker compose config --quiet
• real local private and group smoke runs: 20/20 connections, 499/499 ACKs, zero send failures

The checked-in smoke reports validate the harness and are explicitly labeled as local development results, not production capacity claims.

Closes #37