Bootstrap orchestrator core: scope -> architect -> build -> full-fledged loop

[suleimansh]

Child of #116 (bootstrap). The spine.

Sequence the existing primitives into the bootstrap flow:

• Scope (one upfront question: prototype vs full + intent; the only interrogation).
• Architect: agent picks the stack (Vike + universal-orm), narrates it, records key choices to the decisions ledger (Decisions bookkeeping — stop the AI re-pitching rejected ideas #112). No permission asked.
• Build + narrate: execute the plan in LocalRunner (ai-autopilot: LocalRunner — first real runner adapter (host fs + child processes) #106) via the stack personas (ai-autopilot: personas + prompts library (stack-aware knowledge) #98), streaming narration over the surfaces transport (ai-autopilot: surfaces (terminal / in-page / background) #100); user can interrupt (AbortSignal).
• Full-fledged loop: fire major-change / ui-flow events into the loop ("The loop" — event-triggered prompt chains (major change → review/quality/security; UI flow → QA/UX) #113), repeating with fresh context until the production-grade checklist (Production-grade checklist prompt + { blockers } verdict convention #121) reports no blockers or a maxPasses budget stops it.

Offline-testable with fakes (fake runner + stub agents). Depends on #120 (generic surface) and #121 (checklist verdict).