feat(job): move metadata storage to a database

[berstpander]

Summary

Move Job and Group metadata from OSS to a database-backed SDK and decouple the local CLI from persisted metadata.

Closes #1306

Changes

Metadata SDK

• Add database models for Group and Job metadata.
• Use UUID group_id and job_id as canonical identifiers.
• Add transactional create and update operations.
• Support complete Job lookup by job_id.
• Support namespace Group listing.
• Support Group Job filtering by status, category, task, and Job type.
• Add cursor pagination, Group statistics, and combined Group details.
• Accept a caller-owned SQLAlchemy session factory so connection credentials remain in the trusted metadata service.

Job identity

• Generate an independent UUID for every Job.
• Allow callers to provide an existing job_id.
• Propagate the same ID through SDK results, labels, environment variables, and JSONL events.
• Remove process-local run_id behavior.

CLI

• Remove dependencies on persisted metadata repositories.
• Keep --task, --tasks, and --all.
• Add optional --job-id for single-job execution.
• Return the effective job_id in JSONL Job lifecycle events.
• Replace metadata-based resume with explicit sandbox, process, and session parameters.
• Keep OSS access for dataset discovery and execution artifacts.

Compatibility Notes

• Remove legacy Run metadata commands and Run ID lookup.
• Remove ROCK_RUN_ID, rock_run_id, and JSONL run_id.
• OSS remains responsible for datasets and execution artifacts.
• The local CLI does not access the metadata database.

Testing

• SDK Job: 329 passed
• CLI Job: 35 passed
• Legacy SDK Agent Job: 27 passed
• Ruff and format checks passed