Cluster Topology: address #474 review comments on memory basis and failover roles, plus node row UI updates

[nassery318]

Two review comments from @sarthakaggarwal97 in discussion #474, plus some frontend UI updates to the Cluster Topology view (#395).

1. Don't use total_system_memory as capacity

• Only count memory toward utilization when the node has maxmemory set.
• Cluster Memory card: show ∞ for capacity when any node has no maxmemory, instead of adding host RAM to the total.
• Nodes without maxmemory keep their host-RAM percentage in the tooltip, marked as not counted, with a dashed badge. They no longer count toward the High badge or "Nodes
  Flagged".

2. Keep node roles in sync after a failover

• Re-read the cluster topology (CLUSTER SLOTS) on each 5s poll and send it with the cluster data, so primaries and replicas regroup after a failover.
• When the topology changes, update the server's node registry and reconcile metrics servers.

3. Frontend UI updates

• Show host:port as the node row label instead of server_name, and drop the separate host:port line.
• Replica rows use a smaller label than primary rows.
• Remove server_name from node search and update the search placeholder to "Search nodes by host or port...".

Why?

• Memory: If several nodes share a host, adding total_system_memory for each one counts the same RAM more than once. That inflates cluster capacity and flags nodes as High against a limit they don't have.
• Failover: Roles are fixed at connect time. After a failover, the demoted node keeps its PRIMARY label and the promoted node stays a REPLICA, which hides its metrics and leaves it out of "Nodes Flagged".
• Node rows: host:port identifies a node in the cluster without ambiguity, and it matches how nodes are addressed everywhere else in the view.