Promote kubeconfig flags (--kubeconfig/--context/--namespace) to persistent on root

[saadqbal]

Problem

Today tracebloc cluster info redeclares --kubeconfig, --context, and --namespace as local flags. As we add more cluster-talking subcommands (dataset push, dataset list, future ingest status), each one will need the same three flags. Redeclaring per-command means:

• The same flag-declaration code gets copy-pasted 4+ times
• Help text drifts as docs are tweaked per command instead of in one place
• Default values can silently diverge

Fix

Promote --kubeconfig, --context, --namespace to persistent flags on either the root command or a clusterFlags middleware. Subcommands inherit them automatically.

Implementation sketch (internal/cli/root.go):

type GlobalFlags struct {
    Kubeconfig string
    Context    string
    Namespace  string
}

func NewRootCmd(info BuildInfo) (*cobra.Command, *GlobalFlags) {
    flags := &GlobalFlags{}
    root := &cobra.Command{...}
    root.PersistentFlags().StringVar(&flags.Kubeconfig, "kubeconfig", "", "...")
    root.PersistentFlags().StringVar(&flags.Context, "context", "", "...")
    root.PersistentFlags().StringVarP(&flags.Namespace, "namespace", "n", "", "...")
    // Subcommands read flags via the struct rather than redeclaring.
    root.AddCommand(newClusterCmd(flags))
    ...
    return root, flags
}

Pass the flags struct through subcommand constructors so each can call cluster.Load(cluster.KubeconfigOptions{Path: flags.Kubeconfig, ...}) consistently.

Acceptance criteria

• tracebloc --kubeconfig X cluster info and tracebloc cluster info --kubeconfig X both work
• Same for --context and --namespace / -n
• Adding a new subcommand that talks to the cluster (Phase 3's dataset push) requires zero per-command flag-declaration
• Existing cluster info unit tests still pass

Found by

Self-review of PR #2 (Phase 2 cluster discovery). Filed as v0.2 polish so it doesn't gate the v0.1 push-only MVP.