feat: agent-friendly threat-model authoring end to end

[Hacks4Snacks]

feat: agent-friendly threat-model authoring end to end

What & why

TMForge could author .tm7 models verb-by-verb, but only by GUID and every add returned a
fresh GUID that changed on every rebuild, so any non-trivial model had to be authored as a
GUID-capturing shell script, and no companion doc or report could cite a stable id. Authoring was
also imperative-only: the diagram diffed poorly and there was no reviewable source of truth. This
threads ergonomic, reviewable authoring through the CLI stable references, a declarative manifest,
logical boundaries, auto-layout, and better discoverability driven by real agent-authoring feedback.

Highlights

• Aliases & stable ids: add --alias P1 gives an element a durable handle and a deterministic
  id (the same alias → the same GUID across rebuilds, so docs/reports can cite it). connect / set /
  remove / rename / show now resolve GUID, alias, or unique name.
• Declarative manifests: tmforge apply <manifest.json> builds a whole model in memory and writes
  it atomically (all-or-nothing, --dry-run, idempotent re-runs); tmforge export emits a
  diffable JSON manifest from any model. A model becomes reviewable model-as-code in a PR.
• Logical boundaries & layout: add --boundary <ref> records trust-boundary membership and places
  the element inside it (so crossings are computed and membership round-trips through export); new
  tmforge layout applies a deterministic layered auto-layout so authors never hand-place coordinates.
• External-entity auth: external entities gain an AuthenticationScheme property; TM1023 clears
  when an external declares a scheme (e.g. a token or public key), so the finding is resolvable in-model.
• Discoverability: properties --explain maps each property value → the rule + severity it
  triggers (predict lint before running it); new tmforge schema documents the --json envelope and
  every command's data shape, so automation isn't written against probed output.
• SVG export: report --format svg writes a standalone diagram (all pages stacked) for PR review,
  distinct from the HTML report.
• Command registry: dispatch, --help, and schema are now generated from a single
  CommandCatalog, so adding a verb updates all three at once and can't drift out of sync (test-guarded).
• Docs: README, overview, and cli-reference updated for the new verbs/flags.