node_cpu: document an invocation that runs - #99

Follow-up to #96, from its review.

python -m mote_bringup.tools.node_cpu — the form node_cpu.py's own docstring
gave — does not run:

$ pixi run -- python -m mote_bringup.tools.node_cpu --help
ModuleNotFoundError: No module named 'mote_bringup.tools'

mote_bringup/tools/ has no __init__.py, so setup.py's find_packages()
never picks it up. That is deliberate: these are harnesses run from a checkout,
not code the robot installs, which is why every sibling tool (slip_replay,
icp_excursions, odom_health, icp_gate_replay, camera_layer_decay)
documents the path form. node_cpu was the only one claiming otherwise, and the
tool itself was always invoked correctly during the measurements — by path on the
Pi — so nothing exercised the documented form.

The review found two occurrences in node_cpu.py; there were four. The same
broken form was in docs/tuning/2026-08-11-monitor-cpu/runs.txt, and the tuning
doc carried a bare mote_bringup/tools/node_cpu.py --summary <csv>, which is not
a runnable command either.

All four now use pixi run node-cpu — the task #96 already added, and the form
camera_layer_decay documents. Pixi appends trailing arguments, so
--summary/--nodes/--duration pass straight through; verified by
re-deriving the figures the tuning doc quotes:

$ pixi run node-cpu --summary docs/tuning/2026-08-11-monitor-cpu/probe-tf.csv
{ "samples": 90, "span_s": 89.1, ... "cpu_probe_floor": {"mean": 0.5, ...} }

The alternative fix — adding mote_bringup/tools/__init__.py so -m works — was
rejected: it would make find_packages() ship the tools into the install space
and onto the robot, a packaging change nobody asked for, and would leave
node_cpu the odd one out among its siblings in the opposite direction.

Docs only: 4 lines across 3 files, no behaviour change. pre-commit run --all-files clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01X6o6x5Nn7QYUMxg2F3Wy2t