fix: extend realtime OpenClaw consult wait

[rogerchappel]

Summary

• Extend CrewCMD's realtime OpenClaw consult wait from 110 seconds to a 10 minute bounded window and raise the route max duration accordingly.
• Recover empty or missed realtime final events by reading durable chat.history, preferring the assistant row matching the OpenClaw run id/idempotency key, then falling back to the consult prompt.
• Add a regression test for an empty final event recovered from chat history.

Investigation

OpenClaw docs describe talk.client.toolCall as the browser realtime path for openclaw_agent_consult: clients receive a run id, wait for normal chat lifecycle events, then submit the provider-specific tool result. The Control UI docs also distinguish live chat delivery events from durable chat.history, and say history is reloaded after tool-final events.

CrewCMD's relay route followed the event-wait pattern, but it added a hard 110 second timeout and returned OpenClaw completed without returning text. when the final event had no text. That explains both failure modes reported here: long OpenClaw runs crossed CrewCMD's local timer, and final events without inline text were not recovered from the durable transcript.

Verification

• pnpm vitest run 'src/app/api/runtimes/[id]/talk/realtime/relay/route.test.ts'
• pnpm typecheck
• pnpm exec eslint 'src/app/api/runtimes/[id]/talk/realtime/relay/route.ts' 'src/app/api/runtimes/[id]/talk/realtime/relay/route.test.ts'
• git diff --check
• pre-push hook: pnpm typecheck && pnpm build passed. Build emitted existing Turbopack NFT warnings around next.config.ts / openclaw-config-parser.ts.

Risk

Medium. This keeps an HTTP tool-call request open longer, matching the current synchronous relay architecture. A follow-up background consult handoff would be better for truly unbounded OpenClaw runs.