Add Evo Studio autonomous cloud training and dataset workflow

[pearlq12345]

Summary\n\nThis is the all-in-one upstream integration PR for the RoboClaw / Evo Studio autonomous training work currently staged in the fork stack. It brings together the cloud training bridge, account/credit ledger, dataset upload and exchange flow, provider/chat reliability fixes, cloud supervisor auto-repair, and route/module cleanup.\n\nIncluded highlights:\n- Add EVO_Train cloud bridge and VLA/RL backend profiles.\n- Add account balance, credit points, cloud training holds, settlement, and dataset reward plumbing.\n- Add dataset adapter contract, dataset push CLI, upload completion, storage/auth refs, and dataset UI improvements.\n- Add cloud training planner/start/status/current/artifacts/resources APIs.\n- Add cloud supervisor with safe_auto same-runtime repair, bounded retries, training-time OOM intervention, and artifact/metric extraction.\n- Add local SSH runtime binding/rebinding flow for existing AutoDL/SeetaCloud instances.\n- Add provider settings and chat connection reliability improvements.\n- Split large route/CLI/curation helpers and surface previously hidden operational errors.\n- Keep train_cloud.py below the 1000-line target by moving artifacts, policy, SSH binding, and watch-loop logic into focused modules.\n\n## Notes\n\nThis PR is intentionally broad because it represents the current complete Evo Studio product stack. For easier review, the same work is also organized as fork-side stacked PRs:\n- pearlq12345#3\n- pearlq12345#4\n- pearlq12345#5\n- pearlq12345#6\n\n## Verification\n\n- python -m pytest tests/test_evo_train_routes.py -q -> 82 passed\n- python -m pytest tests/ -x -q -> 643 passed, 234 skipped\n- cd ui && npm run build -> passed\n- git diff --check -> passed\n

[pearlq12345]

Closing for now: opened before explicit owner confirmation. The changes remain available in the fork/stack for review and can be re-opened or re-submitted after confirmation.