fix(kiloclaw): avoid controller probes for stopped runtimes

[pandemicsyn]

Summary

• Add a Durable Object-level guard so gateway/controller RPCs fail fast unless the KiloClaw instance state is running, preventing stopped runtimes from being probed through provider routing.
• Keep morning briefing status polling graceful by treating the new non-running guard as a warm-up/unavailable state instead of surfacing repeated errors.
• Tighten the admin instance detail polling gate so gateway status and controller version queries only run when the DO-reported runtime state is running.

Verification

• stopped instance, messed with admin ui
• stopped instance, cruised around user ui

Visual Changes

none

Reviewer Notes

• The main behavior change is architectural: controller RPC safety now lives in requireGatewayControllerContext, so future controller-backed UI paths inherit the stopped-runtime protection automatically.
• waitForHealthy() is intentionally unchanged because lifecycle/start reconciliation still needs to probe while a runtime is starting.
• User/admin controller mutations now return a 409 conflict when the DO state is not running instead of attempting provider routing.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (5 files)

• apps/web/src/app/admin/components/KiloclawInstanceDetail.tsx — no issues
• services/kiloclaw/src/durable-objects/kiloclaw-instance.test.ts — updated test correctly inverts expectation to match new stopped-runtime guard
• services/kiloclaw/src/durable-objects/kiloclaw-instance/gateway.test.ts — no issues
• services/kiloclaw/src/durable-objects/kiloclaw-instance/gateway.ts — no issues
• services/kiloclaw/src/routes/platform.ts — no issues

---

_{Reviewed by claude-4.6-sonnet-20260217 · 1,040,105 tokens}