init: pi_coder_v2 — Single-Server (Option B)

Beide Rollen (Coder + Judge) auf einem llama.cpp-Server (Port 8001).
switchModel wechselt zwischen qwen3.5-coder und qwen3.5-judge via
llama-cpp-single-Provider ohne Server-Neustart. start-single.sh startet
den gemeinsamen Server, GPU wählbar per GPU_DEVICE-Env-Variable.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

start-servers.sh

@@ -0,0 +1,30 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+LOG_CODER=$(mktemp /tmp/coder_XXXXXX.log)
+LOG_JUDGE=$(mktemp /tmp/judge_XXXXXX.log)
+
+echo "[*] Starte beide Server parallel ..."
+bash "$SCRIPT_DIR/start-coder.sh" > "$LOG_CODER" 2>&1 &
+PID_CODER=$!
+bash "$SCRIPT_DIR/start-judge.sh" > "$LOG_JUDGE" 2>&1 &
+PID_JUDGE=$!
+
+wait_result() {
+  local PID="$1" NAME="$2" LOG="$3"
+  if wait "$PID"; then
+    echo "[✓] $NAME bereit"
+  else
+    echo "[✗] $NAME fehlgeschlagen — Log:"
+    cat "$LOG"
+    return 1
+  fi
+}
+
+RC=0
+wait_result "$PID_CODER" "Coder (:8001)" "$LOG_CODER" || RC=1
+wait_result "$PID_JUDGE" "Judge (:8002)" "$LOG_JUDGE" || RC=1
+
+rm -f "$LOG_CODER" "$LOG_JUDGE"
+exit $RC