Publish CVML Playbook

[danielholanda]

Description

This PR publishes the CVML playbook to the official AMD Playbooks website.

[danielholanda]

@adamlam2-amd
Can you take a look?

[sreeram-11]

@danielholanda,

FYI:

1. Enabled Linux
2. Enabled STX and KRK as a supported_platform, tested_platform, and published_platform

[sreeram-11]

• Installed all dependencies listed in the README on STX and KRK runners (Windows and Linux), effectively setting up these machines to run the playbook.
• On Linux Halo runners, removed the existing OpenCV installation and reinstalled it using the commands provided in the README.
• With the updated installation method, OPENCV_INSTALL_ROOT is now set to /home/user/build/install across all Linux runners.
  • CI tests have been updated accordingly to reflect this change.

[sreeram-11]

@anna-amd-com. @justichu-amd

NPU driver/runtime seems to be a mandatory requirement, not optional. The current GPU fallback statement is not reliable and should be revisited.

Observed behavior:

• In all cases, CVML selects:
  • [Face Detector] Using ONNX engine, NPU backend
• This forces face detection to run on the NPU, with no fallback to GPU.
• When the NPU driver is not installed, the workflow fails.
  • CVML does not fall back to GPU as expected.
  • Fails with a segmentation fault when NPU drivers are missing.
  • Example failing run: https://github.com/amd/playbooks/actions/runs/27188370389?pr=301
  • After installing the specified NPU driver versions, tests pass successfully.

Conclusion:

• NPU drivers are effectively required for these environments.

Recommendation:

• Update the documentation to mark NPU driver installation as mandatory.
• Remove or correct the GPU fallback statement to reflect actual behavior.

Please confirm and advise.

[adamlam2-amd]

pending further approval and instruction.