fix(ai-evals): split compound pipeline delegation assertion into two

[orionnye]

Summary

• Split the compound assertion "should build a self-contained Task prompt with the pipeline file path and return expectation" into two separate assertions: one for file path inclusion, one for return expectation
• The original assertion was failing at 2/4 runs — splitting it makes it easier to diagnose which part is actually failing

Test plan

• Run pipeline-skill-test.sudo eval and verify both new assertions are evaluated independently

Made with Cursor