Multi-cluster dashboard: cluster nodes aren't grouped together in the node list

[Austin519]

Running the multi-cluster trusted-SSO beta (#114) in production with a single Companion fronting six independent clusters.

In the node list, a cluster's Primary and Secondary don't render adjacent — the Primary sorts to the top while its Secondary appears lower among the other secondaries, so one cluster's two nodes end up split across the page. With several clusters this makes the view hard to scan (e.g. our core cluster's node #2 is nowhere near node #1).

Grouping nodes by cluster (or sorting group → role within each group) would keep each cluster's nodes together.

Not blocking. (Split from #125.)

[Austin519]

Dug into the mechanism on our side, in case it's useful.

The scatter is driven by node→cluster-role matching. The companion appears to match each configured node against the cluster's advertised node identity to decide Primary vs Secondary. When a node is reached via a proxy hostname that differs from the cluster's advertised identity, that match fails and the node falls back to "Secondary" — then the primaries-first ordering splits the pair.

Concretely, across our deployment:

• Core cluster configured by the same internal IPs the cluster advertises (192.168.20.2/.3) → roles detected correctly (Primary + Secondary), nodes stay paired.
• Remote sites configured via Cloudflare hostnames (dns1.<zone>.com) while the cluster advertises dns1.cluster.<zone>.com at an internal IP → no match → both nodes show as Secondary, and the pair scatters.

So beyond the sort itself, grouping the list by the configured group id (rather than by detected role) would keep each cluster's nodes together regardless of whether role detection succeeds. Happy to test any change.