report why a Node is left out of the target groups

[m1ron0xFF]

What

Improves the report about Nodes that target group synchronization leaves out.

Says which of two causes an empty spec.providerID is. The two need completely
different remediation, and skipping the Node without naming the cause turns a real
capacity loss into a silent one. The node.cloudprovider.kubernetes.io/uninitialized
taint tells them apart at no cost — it sits on the Node object, so no cloud API call
is involved:

Node state	What it means	Reported as
Taint present	kubelet did hand the Node to a cloud provider and cloud-node has not initialized it yet, normally a matter of seconds	awaiting cloud-node initialization
Taint absent	the Node was never offered to a cloud provider: either a genuine static Node, or a cloud VM whose kubelet is missing --cloud-provider=external	no <taint> taint, never handed to a cloud provider

Simplifies the bookkeeping behind that report. lastSkippedNodes goes from a
mapset.Set to the rendered string, so the accumulation loop, the nil guard and a
duplicate strings.Join disappear, and the zero value means "nothing skipped before"
on its own. Reasons are sorted, which is what makes the comparison correct — Node
order from the informer is not stable — and also keeps the logged line stable between
reports.

hasTaint is shared with nodeEligibleForLoadBalancer, which carried the same loop.

Why not backfill the ProviderID instead

#195 hit the second case in production and proposed resolving the Instance by Node
name and patching spec.providerID. That was not taken:

• spec.providerID is immutable once set, and the value would come from a Node-name
  lookup, so a wrong match becomes permanent — unfixable without deleting the Node
  object.
• cloud-node already owns this: it sets spec.ProviderID when empty, using the same
  name-based fallback. Doing it again from the load balancer path races that controller.
• The root cause is cluster configuration, which also affects Node addresses, zones and
  instance types — far more than load balancers.

Reporting which cause it is, is the part that belongs in the target group syncer.

Base

Based on fix/skip-empty-provider-id-nodes (#198), not on master: this changes
partitionNodesByProviderID, which #198 introduces. GitHub will retarget this PR to
master once #198 merges.

Testing

go build ./..., go vet ./pkg/..., go test ./pkg/....

TestPartitionNodesByProviderIDDistinguishesUninitializedFromStaticNodes asserts the
two causes produce different messages. TestNewlySkippedNodesReportsOnlyOnChange
covers the dedup, including that reasons are sorted and that a reshuffled slice is
treated as unchanged. Both were verified by mutation — inverting the taint check and
dropping the sort each make the corresponding test fail.