docs: DocxDiff default migration plan (D4 execution)

[JSv4]

Adds a single planning doc — docs/architecture/docxdiff_default_migration.md — laying out how to promote DocxDiff to the blessed default comparison engine and demote WmlComparer to a documented, still-reachable legacy engine (execution plan for decision D4).

Docs-only; no code or behavior change in this PR.

What the plan captures

• Current state — the green/ratcheted parity evidence (179/179 GetRevisions, 39/39 markup, 84/84 consolidate-soundness) and the asymmetric client wiring that blocks a clean flip today.
• Milestones (sequenced, gated):
  • M-A (BLOCKING) — close the four multi-reviewer Consolidate edge-gaps before the flip: note-scope N-way merge, structural table changes, cross-reviewer split/merge, cross-kind note nesting.
  • M-B — add one shared engine selector across WASM / npm / tools/redline, seeded to WmlComparer (the default is not flipped here).
  • M-C — author + run a Microsoft Word manual-verification checklist (artifact doesn't exist yet).
  • M-D — burn-in window with explicit pass criteria.
  • M-E — the flip (default → DocxDiff) + legacy re-labeling in docs and /// remarks only (no [Obsolete] this cycle).
  • M-F — [Obsolete] deferred to a future major.
• Gate exit criteria (G2, D4), four-layer ripple checklist, risk/rollback (the engine selector is the rollback), and a definition of done.

Decisions baked in (per request)

• Consolidate gaps block the flip (not accepted-scope).
• Legacy marking is docs + XML-doc only first; [Obsolete] deferred.
• Engine flag with one shared default; WmlComparer stays the default during the migration — the flip to DocxDiff is M-E, after the gates clear.

Two decision points are deliberately left open in §8 rather than pre-decided: the flip's FormatComparison default (ModeledOnly vs Full) and its semver bump (major vs minor).