Add optional external compute provider support

[Links17]

Summary

This PR adds an optional external compute provider path for training jobs, with a Seeed Cloud provider implementation behind configuration.

The goal is to make LeLab training easier to run for users who have robot data locally but do not have a CUDA-capable workstation. Seeed Cloud is currently available as a free Seeed-hosted compute option for configured users, and this flow is already being used around LeRobot-compatible Seeed hardware such as reBot Arm B601-DM. Seeed's public B601-DM LeRobot guide covers data collection and training in the LeRobot framework: https://wiki.seeedstudio.com/rebot_arm_b601_dm_lerobot/

The implementation keeps that support optional: without provider configuration, the default local and Hugging Face Cloud flows remain unchanged.

What changed

• Adds a provider abstraction for external training runners.
• Adds Seeed Cloud configuration endpoints and frontend controls.
• Allows training jobs to target local, HF Cloud, Seeed Cloud, or a generic external provider.
• Surfaces external job logs, metrics history, checkpoints, and attach/resume-style tracking in the Jobs UI.
• Adds external checkpoint resolution for inference.
• Keeps imported model support compatible with generic model repos after offline policy validation.
• Adds tests for provider discovery, Seeed Cloud config, external logs/metrics/checkpoints, attach routes, recording behavior, and offline policy options.

Review notes

The Seeed-specific integration is intentionally isolated behind the provider interface and local configuration. If no Seeed Cloud token/config is present, users should only see the existing local/HF behavior plus any public provider flavor catalog exposed by configured providers.

I kept the provider interface generic so other hosted compute backends can be added without coupling the job registry to one vendor.

Test plan

• python -m pytest tests/test_compute_providers.py tests/test_seeed_cloud_config.py tests/test_jobs.py tests/test_record.py tests/test_train.py tests/test_utils_hf_auth.py
• npm run build