fix: improve Render deployment config for reliability

[GeneralJerel]

Summary

• Switch agent from pserv to web type (enables health checks via /ok)
• Switch app from Docker to native Node runtime (simpler builds, Render manages process lifecycle)
• Add auto-scaling (1–3 instances, 80% memory / 70% CPU targets) on both services
• Add /api/health endpoint for app health checks
• Pin NODE_VERSION=22, add SKIP_INSTALL_DEPS=true
• Update fromService reference for new web type

Test plan

• Deploy via Render Blueprint and verify both services start
• Confirm /api/health returns 200 OK
• Confirm agent /ok health check passes
• Verify chat functionality works end-to-end

🤖 Generated with Claude Code

[GeneralJerel]

Review

Looks good! A few minor notes for awareness:

1. --no-frozen-lockfile in the build command means the lockfile isn't enforced during builds, which can lead to non-reproducible deploys. Intentional?

2. Scaling on starter plan — worth verifying Render's starter plan actually supports auto-scaling and doesn't silently ignore the config.

3. docker/Dockerfile.app is no longer referenced in render.yaml — consider deleting it if it's not used for local dev.

Otherwise the changes are clean and well-structured. Agent being public is confirmed intentional.