v2.2.0 — Any source format, inline and narrated

Format-aware entry points, so a rich-text host (a live editor, a document view)
gets the same sized, themed NSAttributedString attachment and accessibility
narration for every supported source format — not just Mermaid — without ever
touching the parsers.

• Format-aware MermaidRenderer API. A new DiagramSourceFormat
  (.mermaid, .dot, .dippin, .sqlDDL, .gitLog) drives
  diagram(source:format:) and format-aware image / pngData /
  attachmentString / altText(source:format:), plus diagram-based twins.
  DOT, Dippin, SQL DDL and git-log sources now render inline and narrate exactly
  like Mermaid. (#46, #47)
• Non-Mermaid formats ride the same NSCache path as Mermaid — keyed by
  format tag + source so same-text DOT and Mermaid can't collide — so a live
  editor doesn't re-parse and re-render on every keystroke.

Fixed

• Arrowheads over tinted group boxes (C4, architecture, block) no longer show
  a pale canvas-colored wedge. The head is now a single opaque triangle
  rather than a translucent head painted over a canvas-colored shaft-eraser —
  which removes both the seam the eraser prevented and the wedge it caused. Both
  render paths (CoreGraphics + the platform-free RenderScene) change in lockstep
  so the draw-vs-scene conformance ratchet still holds. (#23)

Docs

• The architecture/pipeline diagrams in the platform notes and package READMEs
  are now Mermaid (```mermaid — rendered natively by GitHub, and by
  MermaidKit itself), replacing the ASCII-art code fences. MermaidKit now
  dogfoods its own renderer in its docs. (#49)

Full changelog: https://github.com/2389-research/MermaidKit/blob/main/CHANGELOG.md