Fix OpenShell evaluation Secret dependency

[kami619]

Summary

Fixes OpenShell evaluation Pods failing at startup when the namespace does not contain an inference Secret. The PipelineRun already passes llm-api-key for calls through the configured LiteLLM endpoint, so the OpenShell step should use that parameter instead of requiring a separate Kubernetes Secret.

Changes

• Removes the inference Secret from the OpenShell step's envFrom and key references.
• Sets INFERENCE_API_KEY and OPENAI_API_KEY from $(params.llm-api-key) while keeping the OpenShell credential references.
• Adds a regression test that checks the step has no inference Secret references and uses the pipeline parameter for both API key variables.

Test plan

• rtk pytest -q tests/test_openshell_pipeline_profile.py — 5 passed.
• rtk ruff check tests/test_openshell_pipeline_profile.py — passed.
• rtk git diff --check — passed.
• Full rtk pytest -q attempted; collection is blocked in the temporary checkout because sqlalchemy and aiohttp are not installed.

Related

Fixes the missing Secret startup failure observed in the OpenShell evaluation TaskRun.

AI-Attribution: AIA PAI Nc Hin R gpt-6-luna v1.0
AI-Interpretation: https://aiattribution.github.io/statements/AIA-PAI-Nc-Hin-R-?model=gpt-6-luna