crewAI 1.15.38 makes gpt-6-luna the default model and adds eval flags

The release makes gpt-6-luna the default model across crewAI. Agents can now read spreadsheet URLs passed in their input, using tools over every row. Eval usage spans can also be tagged with a target and a per-evaluation ID.

On the eval side, crewai eval accepts --no-open for headless use. A run that exits with an error is now counted as run_failed, and the project .env file is created owner-only when eval turns tracing on. The trace-sharing prompt now explains why it asks, and pressing Enter counts as yes.

Other fixes check API keys for openai-compatible providers on the first call and scope memoized crew methods to their instance. crewai run now exits non-zero when the project script fails. The docs also cover required_inputs on GET /inputs and which flow fields count as inputs.

Teams that rely on the default model will get gpt-6-luna after upgrading unless they pin another model, so check model settings first. CI jobs that call crewai run will now see a non-zero exit code when the project script fails, which may change how failures are caught. Code that depended on memoized crew methods sharing results across instances should be retested.