Fix daily-rig-game-gen workflow timeout

[Copilot]

The Daily Rig Game Generator workflow hit its 30-minute limit and produced no safe output. Its debugging-reflection loop runs up to 3 attempts, each spawning nested rig CLI subprocesses (typecheck + full --server execution, itself a separate LLM session) with no bound on how long any single subprocess could run — leaving no margin for a job with several compounding LLM round-trips.

Changes

• Timeout budget: raised timeout-minutes from 30 → 45, in line with other workflows that use similarly nested subprocess/agent loops (daily-rig-claude-compat at 45).
• Loop bound: reduced MAX_ATTEMPTS from 3 → 2 to cap the worst-case loop duration.
• Subprocess guard: added a 10-minute timeout in runRigEntry that SIGKILLs a hung child process and surfaces it as a failed attempt rather than letting it silently consume the entire job budget.

const timer = setTimeout(() => {
  timedOut = true;
  stderr += `\n[timed out after ${SUBPROCESS_TIMEOUT_MS / 60000} minutes]`;
  child.kill("SIGKILL");
}, SUBPROCESS_TIMEOUT_MS);

Recompiled with gh aw compile daily-rig-game-gen --strict and committed the updated .lock.yml.

• Fixes [aw] Daily Rig Game Generator timed out #385