feat(deploy): document a k3d dev loop for the robot Kubernetes rendering

[Minipada]

Summary

• Adds deploy/robot/k3d/, a Kustomize overlay over deploy/robot/kubernetes/robot-pod.yaml
  that pins dc-ros/dc-uploader to the :jazzy tag (this branch's own published tag —
  never :latest, the same convention ROS's own images use: distro codename, not
  :latest) and sets imagePullPolicy: Never, so a disposable k3d
  cluster can run the robot Pod for real (a scheduler, Services, DNS) with no registry
  reachable from inside the cluster.
• No wrapper script: deploy/robot/README.md's "k3d development loop" section documents
  the loop as plain, commented podman run / k3d cluster create / k3d image import /
  kubectl kustomize | kubectl apply -f - commands. podman run --entrypoint true ghcr.io/.../dc-ros:jazzy gets the already-published image into local Podman storage —
  no local build needed just to try the loop; building your own image for local changes
  needs no doc change, just tag it identically. kubectl apply -k itself refuses a base
  outside its own directory, which is why the apply step pipes through kubectl kustomize --load-restrictor LoadRestrictionsNone instead.
• Adds a "Deployment renderings and a local Kubernetes loop" subsection to
  doc/src/dc/setup.md (the mdbook) pointing at that README section.
• Records the Docker dependency (k3d wraps k3s in Docker containers) as scoped to this
  loop alone in CLAUDE.md's Podman-not-Docker section — building and shipping
  dc-ros/dc-uploader is unchanged, and this is explicitly not the production-parity
  check that's kind validation of the three-tier topology with NetworkPolicy #452's kind + NetworkPolicy cluster.

Verification

Ran the full documented sequence twice: once end to end with lightweight stand-in images
(cluster create ~15-20s, image-reload-and-recreate ~10s, teardown ~2s — proves the
mechanics: hostPath volume staging, the Kustomize overlay, the readiness wait, the
delete-and-reapply iteration path), and again against the real published images
(podman run --entrypoint true ghcr.io/minipada/ros2_data_collection/dc-ros:jazzy and
dc-uploader:jazzy both pull and exit cleanly; the overlay correctly renders
image: ghcr.io/minipada/ros2_data_collection/dc-ros:jazzy + imagePullPolicy: Never).

One thing I couldn't finish live: the podman save + k3d image import step against the
real images, which are ~17 GB each — my sandbox has under 4 GB free disk, and that step
needs the image's size again as a temporary tar on top of Podman's own copy. That's a
sandbox limit, not a defect in the commands, but it's a real cost on any machine, so the
docs now call out the image size and that it's specifically the loading step that's
slow, not cluster creation or iterating on the manifest afterward. Also caught and fixed
along the way: podman save refusing to overwrite an existing tar on a repeat iteration,
k3s failing to boot in restricted/rootless container environments without /dev/kmsg
(fixed with the KubeletInUserNamespace kubelet feature gate), and a Markdown line-wrap
in the mdbook page that was rendering as an accidental bullet list.

prek run --all-files --skip build-doc passes, and mdbook build renders cleanly.

Closes #451

https://claude.ai/code/session_011mBacgaX1VbBvWUfK3RQKU

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 71.46%. Comparing base (d9340bb) to head (ea6d7e3).

Additional details and impacted files

@@            Coverage Diff             @@
##            jazzy     #464      +/-   ##
==========================================
+ Coverage   71.44%   71.46%   +0.02%     
==========================================
  Files         124      124              
  Lines        7563     7563              
==========================================
+ Hits         5403     5404       +1     
+ Misses       2160     2159       -1     

Flag	Coverage Δ
cpp-jazzy	71.46% <ø> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.