AI builder eval gains workflow debugging support

When evaluating AI-built workflows, the old system deleted everything after each run — developers had no way to inspect what the AI actually generated. A new --keep-workflows flag lets developers preserve workflows for post-mortem debugging without manually saving them.

The changes also address an API compliance issue: n8n now requires workflows to be archived before deletion. The client library now calls the archive endpoint before hard-deleting, preventing errors in cleanup routines.

Error handling in EvalExecutionService was also improved — execution failures now return proper evaluation results instead of propagating 500 errors.

In the AI builder package (packages/@n8n/instance-ai), the CLI flag flows through the argument parser into the test harness runner. The n8n-client handles the archive-before-delete sequence, and scenario hints now pass through to pin data generation so mock nodes reflect the actual evaluation context.