refactor: use skillimage packages for skill.yaml parsing and discovery

Summary

Replace docsclaw's local skill-related packages with imports from
skillimage to avoid
duplication and keep the two projects aligned.

What to replace

What to keep in docsclaw

• pkg/skills/loader.go — Discover(), LoadContent(),
  BuildSummary() (runtime skill loading, agent-specific)
• pkg/skills/a2a.go — ToAgentSkills(), MergeSkills()
  (AgentCard population, A2A-specific)

These functions would call into skillimage types instead of local
ones.

Skillimage packages available

Steps

1. Add github.com/redhat-et/skillimage as a dependency
2. Replace pkg/skills/card/ types with skillcard.SkillCard
3. Replace ParseSkillYAML() with skillcard.Parse()
4. Update ToAgentSkills() to map skillcard.SkillCard → a2a.AgentSkill
5. Remove pkg/skills/card/ package
6. Verify skill list and Discover() still work

Context

Both projects are maintained by the same team. The skillimage
project owns the skill.yaml schema (currently
skillimage.io/v1alpha1) and should be the single source of truth
for parsing and validation. Docsclaw currently has its own
docsclaw.io/v1alpha1 schema in pkg/skills/card/ plus a loose
SkillYAML struct — both should be replaced by the skillimage
types.

Start by reading pkg/skills/card/, pkg/skills/loader.go, and pkg/skills/a2a.go, then review the corresponding skillimage/pkg/skillcard APIs. Add skillimage as a dependency and trace the existing parsing, discovery, and AgentCard mapping flows. Done means the local card package and loose parser are removed, skill list and Discover() still work, and ToAgentSkills() uses skillimage.SkillCard.