RivetKit: startEngine: false doesn't reach the native bridge, which still spawns an Engine (EngineSpawnMode::Auto) #5926

Written by an AI agent (Claude) on behalf of @jasonkuhrt.

A registry set up with startEngine: false can still start its own Engine. The native config bridge never receives that option.

setup({ use, endpoint, startEngine: false /* , ... */ });

We run the Engine ourselves and register runners against it, so a runner should fail registration when the Engine isn't reachable, not start a second one. In rivetkit-napi's serve_config_from_js (2.3.30), the spawn mode is hardcoded to EngineSpawnMode::Auto (registry.rs, around line 845; around line 706 in 2.3.10). startEngine from the TypeScript config doesn't reach it, and RIVETKIT_ENGINE_SPAWN isn't read on that path either.

So with a loopback or *.localhost endpoint, a /health probe that misses its one-second window takes the start-or-reuse path, and the runner process spawns an Engine of its own, which then outlives the stack that owns the real one. Remote endpoints aren't affected.

Expected: startEngine: false maps to the core's Never mode, so the runner never spawns an Engine, and an unreachable Engine fails the runner's normal connection and registration instead.