[Bug] ray re-init fail will cause OOM

**Bug Description**
ray re-init fail will cause multi ray instance local init, which will cause admin process OOM
**Steps to Reproduce**

**Expected Behavior**

**Actual Behavior**

**Error Logs**

**Environment Information**
- **OS**: [e.g. Ubuntu 22.04, macOS 13.0, Windows 11]
- **Python Version**: [e.g. 3.11.5]
- **ROCK Version**: [e.g. 0.2.0]
- **Installation Method**: [e.g. pip install rl-rock, source installation with uv]
- **Docker Version**: [e.g. 24.0.6] (if using sandbox features)
- **Deployment Type**: [e.g. local, distributed, ray]

**ROCK Configuration**
- **Runtime Environment Type**: [e.g. uv, pip, conda]
- **Sandbox Image**: [e.g. python:3.11, custom image]
- **Resource Allocation**: [e.g. memory=8g, cpus=2.0]

**Component Affected**
- [ ] Sandbox
- [ ] Actions
- [ ] Deployments
- [ ] SDK & API
- [ ] Envhub
- [ ] CLI
- [ ] Performance & Optimization
- [ ] Documentation & Examples

Start by reproducing the Ray re-initialization failure and tracing the local initialization path that creates multiple Ray instances. Confirm the admin process behavior during a failed re-init; done means a failed re-init does not leave multiple local instances running or cause the admin process to run out of memory.

Written by the indexing model from the issue text.