ProgramBench YAML eval should fail cleanly when generated tasks are missing

[bingran-you]

Mirrors Linear ENG-162: https://linear.app/benchflow/issue/ENG-162/bug-programbench-yaml-eval-crashes-when-cacheprogrambench-benchflow

Problem

Running the checked-in ProgramBench eval YAML before generating .cache/programbench-benchflow can surface as a raw FileNotFoundError / traceback instead of an actionable config error.

Expected

bench eval create --config benchmarks/programbench/programbench-gemini-flash-lite.yaml should validate the YAML tasks_dir early and tell the operator to generate ProgramBench tasks first.

Current fix path

Daily scan 2026-07-11 confirmed this is BenchFlow repo-owned. A focused PR is being prepared to validate missing YAML tasks_dir at config load time and include the ProgramBench generation hint for .cache/programbench-benchflow.

[bingran-you]

Daily scan follow-up (2026-07-12): #918 is the active BenchFlow fix PR for this issue and is currently green on GitHub checks.

Current state:

• PR fix(eval): clean Linear issue UX paths #918 closes this issue and includes the ProgramBench YAML missing-tasks_dir guard plus regression coverage.
• CI is green on head ccb8284cd62e37ab898393fd74e07bbd6b805238 (test, pip-audit, manifest-parity, integration-light / rollout-smoke; integration-scope plan/matrix skipped as expected for this scope).
• No additional BenchFlow code change started from today's scan; remaining gate is human review/merge of fix(eval): clean Linear issue UX paths #918.

Labels updated: added reviewing while the fix PR is open.