Fix eval_awareness and broken_env prompts to name required result fields - #120

Summary

We experienced multiple pydantic ValidationErrors on eval_awareness_scanner and broken_env_scanner scan jobs using OpenAI models (via OpenRouter), because each scanner's prompt never asks the judge for a field that its own result schema requires:

• eval_awareness: the schema requires reason, but the prompt never names the reason field
• broken_env: the schema requires success, but the prompt never mentions it

This PR changes only the two prompt suffixes and bumps both scanners to version=1.

Failure mode

_scan_with_retry sends the result schema as a non-strict ResponseSchema (strict is never set) and validates the completion with ResultClass.model_validate_json. Non-strict structured output guides the output format but does not force required keys. Some models follow the prompt wording over the schema.

Observed errors:

ValidationError: reason   Field required [type=missing]

Other scanners from this repository (reward_hacking, internet_leaves_tracks, sandbagging, etc.) never raise the error in practice because their suffix prompts contain instructions similar to those added by this PR.

Verification

• Patching only the prompt suffix to name the missing field makes both previously erroring scanners validate on the first judge attempt
• ruff check, ruff format --check, and basedpyright are clean, and the scanners test suite passes