Add autonomous cloud training workflow

[pearlq12345]

Summary

• add cloud runtime binding helpers and source resolution for VLA/RL training jobs
• add structured training planning, RLinf recipe discovery, and cloud supervisor routes
• wire cloud training into the agent tool surface and Training Center workflow
• cover planning, source handling, task submission, and status reporting paths

Manual validation

• exercised the Training Center cloud workflow from the local web UI
• checked existing SSH runtime binding, cloud plan creation, source resolution, task submission, status polling, and failure messaging
• verified existing SSH runtimes do not require a platform balance hold for repair submissions
• confirmed stale or unreachable SSH bindings surface as user-visible errors instead of silently blocking the workflow

Verification

• python -m pytest tests/test_agent_cloud_training_tool.py tests/test_evo_train_routes.py tests/test_vla_rl_routes.py -q
• cd ui && npm run build

Stack

Depends on #109.