feat: owasp threat dragon threat model import support

[Hacks4Snacks]

Summary

Add bounded OWASP Threat Dragon v2 JSON import across the CLI, HTTP API,
MCP, and Studio's API/WASM engines.

This enables users to bring supported existing models into tmforge without
recreating diagrams or losing authored threat treatment. Unsupported
semantics are rejected rather than silently changed.

Changes

• Register an import-only threat-dragon provider with content-based
  detection, explicit version checks, and bounded JSON/graph processing.
• Import named pages, actors, processes, stores, rectangular trust
  boundaries, and directed page-local flows with stable identities.
• Preserve authored threats as manual entries, including categories,
  mitigation, priority, supported treatment states, scope, and provenance.
• Carry model metadata and imported threat provenance through canonical
  JSON, engine operations, Studio persistence, and .tm7 exports.
• Fix single-page identity loss and manual threats being assigned to
  the first page during canonical conversion.
• Prevent Studio Save from overwriting read-only source files; offer a
  new .tmforge.json filename instead.
• Preserve non-STRIDE categories during threat editing and return HTTP
  400 for malformed imports.
• Add shared-fixture tests and document the import contract.

No new dependencies or separate threat evaluator are introduced.

Scope and Limitations

• Import only: no native Threat Dragon export or lossless round trip.
• Curved boundaries, bidirectional/detached/cross-page flows, fractional
  rectangles, unknown cell kinds, and unmappable threat states are refused.
• Supported states: Open, Mitigated, Accepted.
• Styling, thumbnails, and free-form connector routing are not retained.
• Out-of-scope flags remain informational; they do not suppress tmforge
  analysis. Missing controls are not inferred as present.
• The upstream demo contains curved boundaries and is intentionally
  refused by this initial implementation.