fix(container): use detached checkout for shallow clones

[mariuspruvot]

Summary

• gh pr checkout fails on shallow clones (--depth=1) with gh CLI 2.90.0, breaking all container sessions
• Root cause: git cannot set up tracking branch info from a shallow clone ref
• Fix: add --detach flag -- containers only need files on disk, not a tracking branch
• Also fixes pre-existing ruff format issue in test_sessions_router.py

Found during E2E QA

Full product validation revealed this as a blocking bug -- no sessions could start. All other flows passed:

• Landing page, OAuth, dashboard, settings, skill selector
• Multi-turn sessions (after fix), session stop, session replay
• Error handling (deleted branch), API endpoints, cleanup

Known issues (pre-existing, not addressed here)

1. Fast-failing container race: when a container exits before SSE stream fully processes, mark_completed may not run, leaving session stuck as RUNNING. Cleanup task handles these as TIMEOUT eventually.
2. Flaky webhook dispatcher tests: structlog.capture_logs misses async log output in fresh venvs (passes on cached venvs).

Test plan

• Verified fix with docker run test: gh pr checkout 20 --detach succeeds on shallow clone
• Verified deleted branch still fails correctly (exit 1, error surfaced)
• Full session lifecycle tested: skill select -> first turn -> multi-turn -> stop -> replay
• make lint passes
• make test passes (237/239, 2 pre-existing flaky failures)

[helprs-local-mariuspruvot]

helPRs session created for this PR.

Skill: challenge-me | Open session