feat(chart): PodMonitor for router-proxy metrics

[Tanguille]

What

A PodMonitor for router-proxy pods, scraping the metrics listener #1457 adds.

Why

Fixes #1472

#1457 adds a --metrics-bind-address listener (container port metrics,
default :9090) but nothing scrapes it: inference-podmonitor.yaml selects
inference.llmkube.dev/service on a port named http, and router-proxy pods
carry neither that label nor that port name. Without this, model-router-dashboard.json
still can't populate even once #1457 merges — its own PR body says so and
deliberately left #1427 open for that reason.

How

router-podmonitor.yaml mirrors inference-podmonitor.yaml's shape:

• Selector: inference.llmkube.dev/model-router: Exists, the stable label
  routerProxySelectorLabels (internal/controller/router_common.go:68)
  already puts on every router-proxy pod.
• port: metrics, path: /metrics.
• job: llmkube-router relabeling, so a future RouterProxyDown rule has a
  stable label to match — same pattern the inference/controller monitors use.
• Gated on prometheus.routerPodMonitor.enabled (default false, matching
  inferencePodMonitor and serviceMonitor's existing default).

This PR is chart-only and depends on #1457 for the port to exist; opening as
draft until #1457 merges (port name/number final).

Assisted-by: OpenCode (helped draft this PR; I reviewed the final change and stand behind it.)

Checklist

• Tests added/updated
• make test passes locally
• make lint passes locally
• Commit messages follow conventional commits
• All commits are signed off (git commit -s) per DCO
• AI assistance (if any) is disclosed above, per CONTRIBUTING.md
• Documentation updated (if user-facing change)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[Defilan]

Approving. I checked the two claims this rests on rather than taking the
description's word:

• routerProxySelectorLabels does set inference.llmkube.dev/model-router: mr.Name, so the selector matches every router-proxy pod and nothing else.
• #1457 does add a container port named metrics backed by
  routerProxyMetricsPort int32 = 9090, with --metrics-bind-address :9090.

Selector and port name line up, the relabel to job: llmkube-router matches
what the inference and controller monitors already do, and defaulting
enabled: false is consistent with inferencePodMonitor and serviceMonitor.

One sequencing note: this is strictly downstream of #1457. Merging it first is
harmless since it ships disabled, but until #1457 lands the PodMonitor would
select pods with no metrics port, so anyone enabling it early gets empty
targets rather than an error. I will merge #1457 first.

Good catch on #1427 too. That dashboard has been advertising "renders empty
today" for a while and this is the half that was actually missing.