fix: prevent bridge death on failed backend switch + auto-restart

[ShaerWare]

Summary

• Root cause: When switching LLM backend to vLLM via admin panel, the bridge was killed BEFORE checking if vLLM is available. If vLLM fails (e.g., on cloud-only server), bridge stays dead permanently until service restart.
• Fix 1: Move bridge shutdown to AFTER successful vLLM startup (same pattern already used in _switch_to_cloud_provider)
• Fix 2: Add periodic bridge health check task (every 60s) that auto-restarts the bridge if it crashes

Test plan

• Switch LLM to vLLM on cloud server — bridge should NOT die if vLLM fails
• Kill bridge process manually — should auto-restart within 60s
• Normal bridge chat still works after changes

🤖 Generated with Claude Code