docs: add Hermes system design maps

[chipoto69]

Summary

Adds system design maps for Harness so contributors can see the actual boundaries between the Claude Code plugin, the Harness meta-skill, generated project artifacts, Agent Teams runtime, workspace state, model policy, and future extension points.

Adds a Hermes translation plan that maps Harness concepts to Hermes skills, profiles, delegation patterns, model/provider policy, generated contracts, and validation loops without pretending Claude Code primitives are portable as-is.

Motivation

• Documentation clarity for contributors reviewing Harness architecture and possible runtime adapters.
• No linked issue.

Scope of change

• Skill / meta-skill logic
• Agent template(s)
• Plugin manifest (.claude-plugin/plugin.json, marketplace.json)
• Documentation (README.md, README_KO.md, README_JA.md, docs/)
• CHANGELOG.md
• CI / GitHub Actions
• Tests
• Other (describe):

Tests

• git diff --check passed
• npx --yes markdownlint-cli2 docs/system-design.md docs/hermes-skill-translation-plan.md passed
• Rendered all 11 Mermaid diagrams with @mermaid-js/mermaid-cli
• Local Markdown link/fence sanity script passed
• npx markdownlint '**/*.md' passes
  • Existing repository-wide Markdown lint debt remains outside this PR's scope; the new docs pass lint directly.
• Manual repro of the change works as expected
• New unit / integration tests added (if applicable)

CHANGELOG update

• Yes — added to CHANGELOG.md
• No — this change is not user-visible (docs-only / internal refactor)

SemVer impact

• Patch — bug fix, no API change (fix:)
• Minor — additive, backward-compatible (feat:)
• Major — breaking change (feat!: or BREAKING CHANGE: in commit footer)
• None — docs / chore / refactor / test

Additional notes

• No plugin manifest changes.
• No version bump.
• No runtime behavior changes.
• The Hermes plan is explicitly marked as an adapter plan, not current Harness runtime behavior.