Fix presenter narration for outcome reports and restore Buttonify

[witbrock]

Merge impact: Canonical effect-outcome reports remain complete on screen while speech becomes a short, human synopsis, and enabled Buttonify quick replies appear again in the Conversations UI.

Merge decision: ready — the affected backend, history, and frontend paths pass targeted regression coverage, with no candidate-caused material blocker remaining.

User outcome

Users no longer hear raw Markdown, presenter tags, identifiers, or diagnostic receipts when Von narrates a canonical effect-outcome report. Enabled Buttonify suggestions are again requested and rendered in the Conversations UI.

Material changes

• Keep the full deterministic effect-outcome report as the authoritative screen channel.
• Generate a bounded spoken synopsis from trusted reconciled effect facts without another model call.
• Repair unsafe duplicated/raw legacy narration while preserving owner/shared-history authority.
• Restore the main Conversations UI per-turn Buttonify opt-in while retaining the user preference as the backend authority.
• Exclude the quoted non-authoritative model draft from Buttonify input.
• Keep ordinary prompt-driven narration unchanged and narrowly recognise legacy reports only at their controlled opening.

Evidence

• .venv/bin/pytest -q tests/backend/test_adaptive_turn_service.py — 178 passed.
• .venv/bin/pytest -q tests/backend/test_von_generate_workflow_instances.py tests/backend/test_von_history_backfill_spoken.py — 34 passed.
• npx jest --runInBand tests/frontend/chatTabSpeechPlanning.test.js src/frontend/web/von_interface/static/js/test/chatTab.test.js tests/frontend/chatPresenterModeRequest.test.js — 279 passed.
• npm run lint:frontend:static -- <three changed frontend test/implementation files> — passed with no findings.
• Python and JavaScript syntax checks plus git diff --check — passed.
• Independent release review found no remaining candidate-caused stop-ship issue.

Ship boundary

• Minimum ship criteria: screen/spoken divergence for canonical reports; no raw canonical Markdown or quoted draft reaches default TTS; current and historical paths remain safe; Buttonify opt-in and rendering work; ordinary presenter narration is preserved.
• Stop-ship conditions: false success or scope narration, raw report narration, non-authoritative draft ingestion, lost ordinary spoken channels, or enabled Buttonify remaining suppressed by the missing request opt-in.
• Non-blocking observations: historical reports without structured facts use a conservative synopsis; quick replies are not reconstructed after a history reload because history omits that debug metadata.
• Non-goals: changing explicit Shift+Speak screen-reading behaviour or redesigning Buttonify as asynchronous delivery.