Add connection sizing formula to command-line help

For --traffic-client-connections help, include guidance for
calculating the number of connections needed based on target
rate, average response time, and the peer's max_concurrent_streams
HTTP/2 SETTINGS value.

Also clarify that each connection implies a dedicated I/O thread
with no separate parameter needed on the client side.

Author: testillano

src/main.cpp

@@ -450,11 +450,20 @@ void usage(int rc, const std::string &errorMessage = "")
        << "  linearly (3 connections ~ 4,000-4,500 subs/s on 8-core hardware).\n"
        << "  Requests are distributed round-robin across connections.\n"
        << "  Equivalent to using multiple endpoints but without range splitting.\n\n"
+       << "  Sizing for real servers with latency:\n"
+       << "    connections >= (target_rate x avg_response_time) / max_concurrent_streams\n\n"
+       << "    Where max_concurrent_streams is the peer's HTTP/2 SETTINGS value.\n"
+       << "    Check with: nghttp -v http://host:port/path 2>&1 | grep MAX_CONCURRENT\n\n"
+       << "    Example assuming max_concurrent_streams=100:\n"
+       << "      12,000 req/s with 0.9s avg response time:\n"
+       << "      connections >= 12000 x 0.9 / 100 = 108\n\n"
        << "  Architecture: each connection = 1 dedicated I/O thread that handles\n"
        << "  HTTP/2 framing, response callbacks, and chain continuations (update,\n"
        << "  delete transforms run inline on the I/O thread). The CPS timer and\n"
        << "  initial request preparation run separately (timer thread or worker\n"
-       << "  pool), so adding connections does not affect timer throughput.\n\n"
+       << "  pool), so adding connections does not affect timer throughput.\n"
+       << "  No separate I/O thread parameter is needed for the client side:\n"
+       << "  connections = I/O threads.\n\n"
 
        << "[--traffic-client-worker-threads <threads>]\n"
        << "  Worker pool for client request preparation; defaults to nproc/2 ("