Add Retire Nexusp Backend for K1C 2025

[arlophoenix]

Stacks on #17.

The 2025 runs two Moonrakers against one Klipper: Creality's forked nexusp on :7125 for the touchscreen, and this script's real Moonraker on :7126. Querying the wrong one does not fail, it answers — curl :7125/server/spoolman/status returns Method not found, which reads as "Spoolman was never connected on this printer" and is wrong. That has cost real debugging time twice on the reference unit, and every user who pastes a :7125 command from a Klipper forum hits it.

This adds an opt-in, default-off option that retires nexusp and puts the real Moonraker on :7125, the port the rest of the Klipper ecosystem assumes. The touchscreen is never patched and could not be — vectorp hardcodes http://127.0.0.1:7125 and is a symlink into tmpfs regenerated at boot from an encrypted blob. What we change is which process answers there.

The daemon is not what is load-bearing. Two JSON-RPC methods are: server.files.get_directory_ex, which drives the file browser's paging, sorting and search, and server.history.count, which the screen calls the moment it connects. A small Moonraker component implements both. Its behaviour was measured against the real nexusp before it was switched off, and once a user retires theirs those measurements cannot be re-derived without reviving it — so the tests ship beside the component as the executable record. Several of the rules are unguessable: get_directory_ex is websocket-only and 404s over HTTP, order arrives as a comma-separated triple, since/before are accepted and ignored, and only .gcode is listed even though Moonraker itself treats .g and .gco as gcode.

The print histories are merged in both directions, because the two daemons keep separate databases. Forward at retirement, so the screen does not lose everything printed before this script was installed — on the reference unit that was 22 of 42 records existing in exactly one place. Backward at restore, so it does not lose everything printed while retired. Merged rows are renumbered into start_time order: Moonraker pages history with ORDER BY job_id, so appending older prints without renumbering presents them as the newest.

This also corrects two comments in #17's file whose premise this PR overturns, and fixes files/services/S50nginx, whose reload path ran nginx -s reload with no -c — so it opened a config that does not exist on this board, failed, and left the old config live while reporting nothing.

Reviewers should know the component reaches into three Moonraker internals: file_manager._list_directory, file_manager._convert_request_path, and raw SQL against history.history_table. All three are checked when the component loads and fail by name, because install_moonraker_nginx runs git checkout master; git pull and every user's Moonraker is therefore a moving target. Note also that Moonraker swallows optional-component load failures and keeps serving, so the option verifies the component appears in /server/info rather than trusting that the port answers.

Test plan

• python3 -m pytest -q — 147 tests, no printer or Moonraker needed (122 component and merge, 25 shell option)
• bash -n clean on every changed shell file
• Config round-trips verified against the real moonraker.conf and nginx.conf this repo ships
• Manual test pass (see checklist below)

There is no CI in this repository, so nothing runs the above automatically.

Manual test pass

• Retire on an idle printer — the screen reconnects on its own, and the file browser pages, sorts, searches, deletes and starts a print
• Upload a fresh gcode after retiring — a thumbnail appears on the screen. This path has never run: every file in the reference corpus already had thumbnails written by nexusp beforehand
• Run "Install Moonraker and Nginx" on a retired printer — the port stays 7125, [creality_compat] survives, and the screen keeps working
• Run "Remove Moonraker and Nginx" on a retired printer — it refuses and points at Restore
• Run a long print end to end on the new arrangement
• Cold-boot — Moonraker comes up on 7125 and the screen reconnects without help
• Restore — the screen returns to nexusp and its history includes prints made while retired

Not addressed

• Thumbnail generation needs Pillow, which is not in the Moonraker this repo ships. moonraker.tar.gz is a Python 3.8 venv with no PIL, and install_moonraker_nginx never pip-installs. The option offers to install it and explains that doing so also repairs Moonraker's own embedded-thumbnail parsing for every user, retired or not. Without it, the disk-union listing still works — that is the part that fixed 66 blank thumbnails on the reference unit.
• printer.info and printer.objects.list answer -32601 for about four seconds after a cold boot, while the screen polls. They are genuine Moonraker methods registered dynamically when Klipper connects. Deliberately not shimmed: register_endpoint returns early for an already-registered path when the incoming registration is remote, and Klipper's are remote, so a static stub would not collide with Klipper's later registration — it would silently win it for the life of the process. A stub answering printer.info forever is worse than a four-second gap.
• server.history.debug.job and server.debug.status are not shimmed. Neither is screen-facing; each was logged exactly once, at connect, with nothing visibly broken. The second emits malformed JSON.
• The "not applicable" marker in the Information menu sits three columns off the tick rows. info_line pads on ${#status}, which counts ANSI escape bytes; Add Disable Creality Stock Services for K1C 2025 #17's existing tri-state marker has the same property. Fixing it means changing shared padding used by every model's menu, which does not belong here.