fix(bench): isolate Terminal-Bench venv

[drewstone]

Summary

• moves Terminal-Bench imports, dataset loading, and tb run execution to an isolated TERMINAL_BENCH_VENV / .venv-terminal-bench
• mirrors the existing COMMIT0_VENV pattern so AppWorld can keep its Pydantic-1 shared env while Terminal-Bench/LiteLLM use Pydantic 2
• adds an offline fail-loud adapter test for the missing isolated venv path

Why

Agent-lab R139 reached the Terminal-Bench adapter but failed preflight before scoring: shared bench/.venv had Pydantic 1.10.26 for AppWorld, while Terminal-Bench/LiteLLM require Pydantic >=2.10. This unblocks the public shell-task readiness lane without any lab-side special case.

Refs #413

Verification

• pnpm install --frozen-lockfile
• pnpm --dir bench install --frozen-lockfile
• pnpm --dir bench exec tsx --test src/benchmarks/terminal-bench.test.mts
• pnpm --dir bench exec tsx --test src/benchmarks/terminal-bench.test.mts src/benchmarks/commit0.test.mts src/benchmarks/appworld.test.mts
• pnpm --dir bench exec tsc --noEmit --pretty false
• pnpm typecheck
• pnpm test
• pnpm build

[tangletools]

✅ Auto-approved drewstone PR — e647c27f

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-06-29T23:02:23Z}