Use text payload for realtime consult results

[rogerchappel]

Summary

• align CrewCMD final realtime consult tool results with OpenClaw Gateway relay manager payloads
• submit final openclaw_agent_consult output to the provider as { text } instead of { result }
• keep the CrewCMD finalText response unchanged for message persistence

Research notes

OpenClaw docs only specify talk.session.submitToolResult as the relay completion mechanism, but OpenClaw’s own Gateway relay manager uses { text } for final realtime consult answers in manager.runtime. The Control UI browser-owned path uses { result }, which appears less appropriate for Gateway-owned relay spoken output. Recent logs show OpenClaw did read /Users/roger/Developer/product-videogen/README.md and returned valid answers, while the spoken realtime output remained poor after final tool.result.

Verification

• pnpm test -- 'src/app/api/runtimes/[id]/talk/realtime/relay/route.test.ts'\n- pnpm typecheck\n- git diff --check\n- pre-push: pnpm typecheck && pnpm build\n