Migrate hyperloom-workload-optimizer skill to AMD-AGI/Hyperloom product repo

[danielholanda]

Description

This PR migrates hyperloom-workload-optimizer skill from amd/skills to AMD-AGI/Hyperloom product repo. The skill will continue to be part of the amd/skills catalog.

This enables AMD-AGI/Hyperloom to be the single source of truth (SSOT) for the hyperloom-workload-optimizer skill.

From now on amd/skills will simply mirror any changes made to this skill instead of requiring code to be changed at amd/skills.

[danielholanda]

Looks like your linter is failing on the skill your team created on amd/skills. Feel free to make changes to this PR directly.

[johnl-amd]

The only red check is ruff, and it is formatting rather than anything
functional: two files want lines rejoined under this repo's 120-char limit.

uvx ruff@0.15.20 format skills/hyperloom-workload-optimizer/

That is 5 insertions and 16 deletions across scripts/preflight.py and
scripts/tests/test_preflight.py, and ruff check --force-exclude passes
cleanly afterwards. Verified on this branch.

Worth using ruff directly rather than pre-commit run here: the hook set needs
python3.11 and fails to build its virtualenv on a machine without it, whereas
ruff is a standalone binary and picks up [tool.ruff] from pyproject.toml
either way.

Everything else is green, and the skill's own checks pass: Structural, Discover
and Results. Behavioral is skipping, which is expected given machine.yml asks
for [mi300x, gpu, rocm] and that lane is dispatch-only.