docs(framework): add plugin concern taxonomy and skill placement rules

[alexsoyes]

🎯 What & why

Neither doc said which plugin a new capability belongs to - the concern taxonomy was only implicit in each plugin.json. This makes placement decidable: a canonical taxonomy plus the rules that follow from it.

🛠️ How it works

• docs/ARCHITECTURE.md:82 - new "Plugin concerns and layers": a concern/layer table for the 6 plugins, then three rules - knowledge vs execution is a firewall, the concern decides placement (not whether a skill exists yet), and orchestration = sequencing across multiple concerns (delegating one sub-step does not make a skill an orchestrator).
• docs/CREATE_PLUGIN.md:7 - new "Adding a skill to an existing plugin": which plugin (by concern, linking the taxonomy) and which number (<NN> encodes logical pipeline order, with the renumbering cost called out).
• CLAUDE.md:19 - points at the taxonomy in ARCHITECTURE.md instead of restating it, so the "never duplicate, link to the canonical home" rule holds.

The taxonomy lives in ARCHITECTURE.md as the single source; the other two link to it. Docs-only - no skill, manifest, or catalog touched.

🧪 How to verify

• Open docs/CREATE_PLUGIN.md:7 and follow the "concern taxonomy" link; it resolves to the new ARCHITECTURE.md section.

✅ I certify

• Docs updated to match the new behaviour.
• I self-reviewed this PR.
• No cross-plugin references introduced.

🤖 Generated with Claude Code