test: end-to-end SWE-bench Docker eval run

[christso]

Objective

Verify the full SWE-bench evaluation pipeline works end-to-end: import from HuggingFace → Docker container → agent solves → code-grader scores → results in studio.

Context

The pieces are in place but have never been tested together:

• PR feat(core): Docker workspace execution environments #971: Docker workspace execution environments (merged)
• PR feat(cli): add HuggingFace dataset import command #986: HuggingFace import command (draft, generates EVAL.yaml with Docker config)
• Issue feat(core): wire base_commit into docker-workspace for SWE-bench evals #987: Wire base_commit into docker-workspace (not yet implemented)

Steps to verify

1. Import a small SWE-bench instance
2. Pull the SWE-bench Docker image
3. Run the eval against a real agent provider
4. Verify: container starts at correct commit, agent works, code-grader runs tests, results in studio

Blocked by

• feat(core): wire base_commit into docker-workspace for SWE-bench evals #987 (base_commit wiring)
• feat(cli): add HuggingFace dataset import command #986 (HuggingFace import PR)

Acceptance criteria

• At least 1 SWE-bench instance runs end-to-end with a real agent
• Code-grader correctly reports pass/fail
• Results visible in studio