docs: add ArghDA vision, architecture & roadmap

Adds docs/arghda-vision.adoc — the wider plan for ArghDA as the
estate's visual proof-wiring / wiring-diagram tool, complementing the
existing MVP spec (arghda-spec.adoc).

Covers:
- The hard boundary: ArghDA organises/visualises/communicates proofs;
  Agda proves (directly or via Echidna). It never proves anything.
- The motivating pain mined from echo-types and absolute-zero
  (orphan modules, hidden postulates, invisible dependency order,
  drifting status tables, refuted-route re-walks, merge drift).
- Three-layer architecture: Rust arghda-core (engine) + AffineScript
  arghda-studio (the visual tool, → typed-wasm/Deno-ESM) + arghda-panll
  (PanLL embedding), with a decision record.
- The honest AffineScript feasibility position and the static-SVG/HTML
  resolution for v1 (pure renderers sidestep the broken interactive
  path; interactive canvas deferred with ArghDA as the forcing function).
- The single richly-tagged proof-node graph data model (status x
  soundness x dependency x glyph x multi-register narrative).
- The full feature set: take-from / visualise / put-back.
- The communication model adapted from nextgen-language-evangeliser
  (RAW/FOLDED/GLYPHED registers, Makaton-style glyphs, 5-slot no-shame
  narrative, audience-parameterised, honest-scope labelling).
- Integration seams: Echidna (Zig API service #3), TypeLL obligations,
  HAR programmatic stage, Groove discovery, kategoria + ideas-to-alphas.
- RSR-compliant shape for the proposed arghda-studio repo and the
  arghda-core work needed to finish the spec.
- A six-phase roadmap and an open-decisions/risks register.

Planning artefact only; no engine or proof code changed.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019GiSiEfgZCte35dyykgBHs

Author: claude