fix(arc_agi_3): exit non-zero when the harness produces no result.json

[ShichengRao]

What does this PR do?

run_solver.main() prints an error when the harness exits without writing result.json — and then falls through to return 0. Any wrapper or CI job reading the exit code sees success for a run that produced nothing. (We hit this integrating run_solver into an offline evaluation rig: a fully failed run looked successful to our automation until we checked artifacts.)

This returns 1 on that path so the printed failure and the exit status agree. The success path is unchanged, so test_offline_e2e's returncode == 0 assertion still holds — and now actually discriminates.

Related issues

None filed; happy to open one if you prefer tracking it that way.

Checklist

• One-line change; python -m ast parses the file (ruff/pytest not run locally — no formatting-sensitive lines touched)
• Tests added/updated (none — existing e2e covers the success path; the failure path has no harness-free test hook)
• No public API or docs impact beyond the corrected exit status
• No new source files

[esarafian]

The production fix remains a single return 1 after the existing cleanup and durable-result check. Added a credential-free, normally collected regression covering missing/valid result.json with harness exit 0/1: all four cases pass; the original source fails both missing-result cases. Valid-result behavior, artifact copy-back, and cleanup are preserved. Ruff lint and format checks pass. Fork CI is approved and running; merge will wait for successful checks. 🤖🤖🤖