feat: add vllm-metrics-tui optional dependency and tmux integration

[samsja]

• Add tui optional dependency group for vllm-metrics-tui
• Update tmux.sh to auto-discover inference server URLs from logs and launch the metrics TUI in a dedicated window

https://github.com/samsja/vllm-metrics-tui

---

Note

Low Risk
Low risk: changes are limited to packaging metadata and a best-effort tmux helper that only runs when vllm-metrics-tui is installed and matching logs are present.

Overview
Adds a new optional dependency extra, prime-rl[tui], sourcing vllm-metrics-tui from Git to enable a metrics terminal UI.

Updates scripts/tmux.sh to optionally create a Metrics tmux window that auto-discovers vLLM inference server URLs by parsing logs/inference/node_*.log (and resolving 0.0.0.0 hosts via orchestrator.log) and then launches vllm-metrics-tui with the discovered endpoints.

^{Written by Cursor Bugbot for commit 49d4aa5. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[cursor]

Hostname resolution returns same host for all nodes

Medium Severity

In multi-node inference where all servers bind to 0.0.0.0 on the same default port (8000), the grep … | head -1 on each iteration of the for node_log loop always returns the same first matching hostname from the orchestrator log. This means every node resolves to the same URL, so METRICS_URLS contains duplicates of a single server rather than distinct URLs for each inference node. The grep has no per-node correlation — it needs to either track already-matched hosts or use node-specific identifiers.