rocminfo fails in native Linux package install CI

[arvindcheru]

Summary
The test_native_linux_packages_install job runs on OSSCI GPU runners but the container only passes --ipc host, so /dev/kfd and /dev/dri are not visible inside the container and rocminfo fails with [WARN] rocminfo failed (may require GPU hardware) even though the overall install test passes.

Steps to reproduce
Run test_native_linux_packages_install.yml on a GPU runner (e.g. linux-gfx942-1gpu-ccs-ossci-rocm).
Complete package install and basic verification (Step 2).
Observe rocminfo output in the job log.

Expected behavior
rocminfo runs successfully inside the container and reports GPU agents (same as other GPU test workflows on OSSCI runners).

Actual behavior
Trying to run rocminfo...
[WARN] rocminfo failed (may require GPU hardware) Install, component checks, RDHC, and pytest still pass — failure is warn-only and subprocess output is not logged.

Root cause
Container options in test_native_linux_packages_install.yml lack GPU passthrough flags used elsewhere (test_rocm_wheels.yml, fetch_test_configurations.py): --device /dev/kfd, --device /dev/dri, GPU group-adds, and OSSCI gha-gpu-isolation-settings env-file.

Proposed fix
Add GPU container options to the test job in .github/workflows/test_native_linux_packages_install.yml, aligned with test_rocm_wheels.yml:

options: --ipc host
--group-add video
--device /dev/kfd
--device /dev/dri
--group-add 992
--group-add 110
--env-file /etc/podinfo/gha-gpu-isolation-settings
--user 0:0

Notes

1. Not a packaging install failure; install validation is unaffected.
2. --privileged was removed in PR6854; explicit device flags are the correct replacement.
3. Verify on all OS profiles (ubuntu2404, rhel8, rhel10, sles16) before merge.