common : avoid side effects around params parsing (#29537)

- register --rpc unconditionally and call llama_supports_rpc() only from its handler
- print server "initialization ..." log after args are parsed

Assisted-by: pi:llama.cpp/MiMo-V2.6-Flash-RL

common/arg.cpp

@@ -2673,16 +2673,17 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.video_ffmpeg_bin_dir = value;
         }
     ).set_examples(mmproj_examples).set_env("LLAMA_ARG_VIDEO_FFMPEG_DIR"));
-    if (params.is_gen_docs || llama_supports_rpc()) {
-        add_opt(common_arg(
-            {"--rpc"}, "SERVERS",
-            "comma-separated list of RPC servers (host:port)",
-            [](common_params & params, const std::string & value) {
-                add_rpc_devices(value);
-                GGML_UNUSED(params);
+    add_opt(common_arg(
+        {"--rpc"}, "SERVERS",
+        "comma-separated list of RPC servers (host:port)",
+        [](common_params & params, const std::string & value) {
+            if (!llama_supports_rpc()) {
+                throw std::invalid_argument("RPC not supported in this build");
             }
-        ).set_env("LLAMA_ARG_RPC"));
-    }
+            add_rpc_devices(value);
+            GGML_UNUSED(params);
+        }
+    ).set_env("LLAMA_ARG_RPC"));
     add_opt(common_arg(
         {"-lm", "--load-mode"}, "MODE",
         "model loading mode (default: auto)\n"

tools/server/server.cpp

@@ -102,12 +102,12 @@ int llama_server(int argc, char ** argv) {
     // touch it. lifecycle is symmetric, stop_gc() runs in clean_up() before backend free
     server_stream_session_manager_start();
 
-    SRV_INF("%s", "initializing ...\n");
-
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_SERVER)) {
         return 1;
     }
 
+    SRV_INF("%s", "initializing ...\n");
+
     llama_backend_init();
     llama_numa_init(params.numa);