fix(cli): guided error for --epochs 0/negative instead of raw traceback

[ptlnextdoor]

Summary

--epochs 0 (or negative) crashes both run and eval-set with a raw traceback instead of the guided SystemExit every other invalid CLI input gets.

replace(task, epochs=...) reruns Task.__post_init__, which raises ConfigError when the count is below 1. Neither _cmd_run nor _cmd_eval_set caught it — same bug class already fixed for config-file kwargs in #47.

Changes

• Adds _apply_epochs_or_exit, mirroring the existing _resolve_or_exit pattern: catches ConfigError and re-raises as SystemExit(str(exc)).
• Both call sites (_cmd_run, _cmd_eval_set) now go through it instead of a bare dataclasses.replace.
• Drops the now-unused local replace imports in both functions.
• Updates test_cli_run_closes_embodiment_when_validation_raises (was asserting the raw ConfigError leaked, which was the bug) and adds test_cli_eval_set_negative_epochs_is_a_guided_error for the eval-set path, which had no coverage before.

Test plan

• uv run ruff check .
• uv run ruff format --check .
• uv run mypy (strict)
• uv run pytest --cov — 724 passed, 100% line+branch coverage
• uv run pre-commit run --all-files
• Manually confirmed inspect-robots run --task cubepick-reach --policy scripted --embodiment cubepick --epochs 0 now exits with the guided message instead of a traceback

Fixes #145

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[jeqcho]

Thanks for this ptlnextdoor, and sorry for the awkward outcome: #146 was opened a day earlier by the contributor who had claimed #145 in the issue thread, so it got priority and just merged. Closing this one as a duplicate.

That said, your PR got real things right that the merged one didn't. _apply_epochs_or_exit mirroring _resolve_or_exit is the shape this code should have, down to the import hygiene (moving replace into the helper, Task under TYPE_CHECKING). The pytest.raises(match=...) exact-message pinning is stronger than the substring asserts that landed. And you correctly spotted that the old close-embodiment test was asserting the bug itself.

I opened #152 to bring exactly those pieces into main, and your branch is most of the way there already. If you want to send it, it's yours; comment on #152 to claim it. Hope to see more from you either way.