docs: fix multi-AIPC deployment guide gaps and align k3s endpoint port

[KerwinTsaiii]

Summary

Follow-up fixes for the 3 Node Mini-Cluster Example guide, based on a real single-machine deployment (one host running PXE controller + single-node k3s server + JupyterHub) taken end-to-end (login + GPU notebook verified).

• Step 1 — add a passwordless root SSH prerequisite (and an ansible_connection: local alternative for the self-managed service machine) so the PXE playbook no longer fails at "Gathering Facts".
• Step 6 — make sudo / kubectl / ansible-playbook usage consistent: run playbooks without sudo, export KUBECONFIG=~/.kube/config before the first kubectl, and drop the duplicate k3s kubectl verification line.
• Step 11 — append the notebook NFS export to /etc/exports instead of /etc/exports.d/*.conf, which Ubuntu 24.04 silently ignores (left PVCs Pending).
• Step 12 — use authMode: "auto-login" (the single-node default) instead of "dummy", whose login form returns 404.
• Endpoints — align the Apache /k3s/ credential endpoint to port 8080 (pxe_web_port) so the PXE controller can co-exist with k3s Traefik / ServiceLB which owns host ports 80/443.

Notes

• The matching role change (pxe_web_port / Apache on 8080) ships separately in the deploy/fix-pxe-config branch PR.

Test plan

• Docs build renders the updated Steps 1, 6, 11, 12 without MyST syntax errors
• Port references all read :8080 for the /k3s/ endpoints

Made with Cursor