Harness: create local agent runtime harness for server and dashboard

[mashenjun]

Context

The server exposes health and metrics endpoints, and plugins have debug logging, but there is no one-command local harness that starts an isolated mem9 runtime, captures logs, runs smoke checks, and tears down cleanly.

A stronger harness would let agents reproduce bugs and validate fixes without relying on human copy-paste across terminals.

Scope

1. Add scripts/dev-harness.sh or make dev-harness.
2. Start server and dashboard on isolated configurable ports.
3. Capture server logs, dashboard logs, health checks, and metrics snapshots.
4. Run local smoke tests where possible.
5. Ensure cleanup handles interrupts and failed runs.
6. Document required env vars and safe defaults.

Acceptance Criteria

1. One command starts a local isolated server/dashboard harness.
2. Harness records logs to a predictable temp directory.
3. Harness verifies /healthz and /metrics availability.
4. Harness can run at least one local smoke check.
5. Harness exits cleanly and tears down child processes.
6. It does not require production credentials for basic operation.

Effort

~200-500 LoC net.