0.1.7 - Fix VIP + Node Targeting for Multi-Node Clusters

[Handfish]

Problem

VIP + node targeting via gRPC metadata wasn't working, causing etcd to show 0/0 members, lifecycle versions showing "unknown", and K8s checks showing "unavailable" on multi-node clusters.

Root Cause

Incorrect gRPC metadata format. We sent node targets as a single comma-separated value ("node1,node2,node3"). Talos expects each node as a separate header value - one entry per node, not one entry containing all nodes.

Before (broken):
  nodes: node1,node2,node3

  After (correct):
  nodes: node1
  nodes: node2
  nodes: node3

Changes

• Fix metadata format in with_nodes() - use "node" (singular) for single node, "nodes" (plural) with append() for multiple
• Add regression tests - 7 new tests to prevent this bug from recurring
• Improve K8s client - target specific control plane node for kubeconfig fetch instead of failing silently through VIP
• Better UI feedback - show kubeconfig source ("via KUBECONFIG" / "via node 192.168.x.x") and actionable hints when unavailable
• Use IPs instead of hostnames for node targeting - hostnames from etcd peer URLs may not be DNS-resolvable in all environments, so we extract and use IP addresses instead

[Handfish]

Pretty positive this closes #14.

Feel free to reopen the issue if it is not resolved.