sGPU Test Scheduling: Global Work Queue

[VeeraRajasekhar]

Description

Design and implementation notes for .github/scripts/run_queue_sgpu.sh, which
replaces the static one-suite-per-GPU sGPU scheduler with a single LPT-ordered
work queue.

Measured result (level 3, gfx950, 4 GPUs): 6935s -> 2643s wall clock (2.62x),
99.3% GPU utilization Test coverage is
unchanged: 272,037 tests executed across the same 68/9/2 report files.

Example Result: https://github.com/ROCm/TransformerEngine/actions/runs/31151497276/attempts/1#summary-92786276067

Fixes # (issue)

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

• Change A
• Change B

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes