Debian compatibility

[RWejlgaard]

Debian 13 compatibility + quality-of-life fixes

Extends support to Debian 13 (Trixie) while keeping Ubuntu fully working. Also improves script reliability
and adds uninstall scripts.

These changes has been tested against 2 VMs running Debian 13, one master and one slave. I also verified that the master setup script still works on a separate VM running Ubuntu Server 26.04 LTS.

Debian 13 compatibility

• Network configuration - replaces the Ubuntu-only netplan with runtime detection: uses netplan on Ubuntu,
  falls back to nmcli (NetworkManager) on Debian. Fails with a clear error if neither is found
• CNI binary path - explicitly sets bin_dir = "/opt/cni/bin" in the containerd config, overriding Debian's
  default of /usr/lib/cni so Flannel and Cilium plugins are found correctly
• /usr/sbin in PATH - Debian doesn't include /sbin//usr/sbin in non-login shell PATH, causing modprobe,
  sysctl, and swapoff to fail. Added at the top of both scripts
• software-properties-common removed - Ubuntu-centric package that isn't used anywhere in the scripts and
  doesn't exist in Debian's repos
• sysctl -p → sysctl --system - sysctl -p reads /etc/sysctl.conf which doesn't exist on minimal Debian
  installs; --system reads all drop-in config dirs correctly on both distros

Bug fixes

• mkdir -p /etc/containerd - apt install containerd creates the directory, so the subsequent mkdir without
  -p would fail on any re-run
• MetalLB --wait removed - the speaker pods depend on the metallb-memberlist secret, which the controller
  creates on first run. Using --wait caused a deadlock; now waits explicitly for only the controller
  deployment before applying the IP pool config

Script repeatability

Re-running a partially failed script no longer errors out:

• kubeadm init / kubeadm join - automatically runs kubeadm reset -f first if a previous run got that far
• helm install for NFS and SMB CSI drivers - changed to helm upgrade --install
• helm repo add - added --force-update to all four calls
• kubectl create secret for SMB credentials - uses --dry-run=client -o yaml | kubectl apply -f -

Uninstall scripts

New uninstall/ directory with:

• cleanup_master_node.sh - tears down everything the master script installs: kubeadm reset, Kubernetes
  packages and apt repo, containerd, Helm, NFS/Samba servers (share directories preserved), Cilium/Hubble/Flux
  CLIs, CNI interfaces and configs, kernel module/sysctl configs, and restores swap from fstab backup
• cleanup_worker_node.sh - same but scoped to what the worker script installs (no Helm, NFS, Samba, etcd, or
  CLIs)