fix(skills): ignore asset markdown in skill_view collision scan

[worlldz]

Fixes #35745

skill_view was treating nested legacy *.md files under another directory skill (for example assets/shared-name.md) as bare-name skill candidates.

That caused false collision reports even when there was only one real skill with the requested name.

This patch keeps the existing collision protection for actual same-name skills, but ignores linked markdown files nested under directory skills when scanning legacy flat skill files.

Validation:
uv run --extra dev pytest tests/tools/test_skills_tool.py -q -k 'asset_markdown_name_does_not_trigger_false_collision or TestSkillViewCollisionDetection'

Result:
7 passed, 80 deselected

[teknium1]

Thanks for the focused regression case. This is already implemented on current main by a broader support-document exclusion.

Automated hermes-sweeper review:

• Commit 9137b86a5286e4ea420e9fb894bce34fee546b3d (fix(skills): ignore support docs in skill discovery) excludes files under references/, templates/, assets/, and scripts/ from skill discovery and collision checks.
• tools/skills_tool.py:1172-1180 applies _is_skill_support_path(found_md) before Strategy 3 records a legacy <name>.md candidate.
• tests/tools/test_skills_tool.py:1228-1261 covers the same nested support-markdown false collision and verifies that the real creative/sketch/SKILL.md resolves.
• The fix shipped in v2026.6.19.

The timeline cross-reference to PR #37164 identified the same root cause. Closing as implemented on main.