feat: Add in-cluster service account support (#768)

[pre]

Motivation

Running k9s inside a cluster currently requires generating a kubeconfig from the mounted service account before starting k9s (wrapper scripts around kubectl config set-credentials etc.). kubectl and client-go fall back to in-cluster credentials automatically; k9s should too.

Addresses #768 (closed as stale — please consider reopening; I'm the original reporter).

Approach

When no kubeconfig and no explicit --server/--token are provided and the service account mount (/var/run/secrets/kubernetes.io/serviceaccount) is present, client.Config synthesizes a single-context kubeconfig named in-cluster:

• server from KUBERNETES_SERVICE_HOST/KUBERNETES_SERVICE_PORT
• CA and token as file references, so token rotation keeps working via BearerTokenFile
• the service account namespace as the default namespace

All downstream context machinery (context activation, per-context config, skins, context view) works unchanged off that synthesized config. Detection mirrors clientcmd's inClusterClientConfig.Possible(), so REST credentials stay consistent with client-go's own in-cluster fallback. An existing kubeconfig always takes precedence, matching kubectl. No new flags.

Testing

• Unit tests: in-cluster detection, kubeconfig precedence, and unchanged no-config behavior (internal/client/config_test.go).
• e2e in kind: pod with a dedicated ServiceAccount and no kubeconfig, k9s started with just the SA mount:

 Context: in-cluster [RW]                          <0> all       <a>       Attach       <ctrl-k>  Kill           … ____  __ ________
 Cluster: in-cluster                               <1> default   <ctrl-d>  Delete       <l>       Logs            |    |/  /   __   \______
 User:    in-cluster                                             <d>       Describe     <p>       Logs Previous   |       /\____    /  ___/
 K9s Rev: v0.0.0                                                 <e>       Edit         <shift-f> Port-Forward    |    \   \  /    /\___  \
 K8s Rev: v1.36.1                                                <?>       Help         <z>       Sanitize        |____|\__ \/____//____  /
 CPU:     n/a                                                    <shift-j> Jump Owner   <s>       Shell                    \/           \/
 MEM:     n/a
┌──────────────────────────────────────────────── pods(default)[1] ────────────────────────────────────────────────┐
│ NAME↑         PF        READY        STATUS          RESTARTS IP                  NODE                     AGE   │
│ k9s           ●         1/1          Running                0 10.244.0.5          k9s-test-control-plane   22h   │
│                                                                                                                  │
└──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘
  <pod>

Startup log:

DBG Using active context context=in-cluster
INF ✅ Kubernetes connectivity OK

(CPU/MEM: n/a is only the missing metrics-server in kind; K9s Rev: v0.0.0 is a dev build without version ldflags — neither is related to this change.)