Add Disable Creality Stock Services for K1C 2025

[arlophoenix]

Creality's stock K1C 2025 firmware runs a telemetry agent, alchemistp, whose only job is to upload printer configuration and logs to Creality. This isn't theoretical — a passive capture on a helper-managed unit caught it resolving devdata.cxswyjy.com and completing a TLS handshake to 98.86.129.225:443, on a printer whose stock cloud blocklist was already active. Alongside it, three more daemons become dead weight once the Built-in Camera Fix is installed: onyxp (cloud WebRTC signalling, which also beacons the printer's serial and MAC at every boot and appends to a never-rotated log every ten seconds — 51.8 MB on the reference unit), thirteenthp (an unauthenticated HTTP media server on 0.0.0.0:8000) and solusp (AI failure detection). All three are blind once mjpg_streamer owns /dev/video0.

This adds a Customize-menu option to disable those daemons and a matching restore, in the same shape as the existing Creality Web Interface pair. The rename uses a disabled. prefix rather than a .disabled suffix, because rcK globs CS??* on this model and a suffix would still match — the same prefix convention the Moonraker service and the Klipper configuration lock already use here.

The WebRTC and AI daemons are gated on the Built-in Camera Fix specifically, not on any camera fix. USB Camera Support filters /dev/video0 out of its device list, so a USB-only install leaves the built-in camera free and those daemons still serving it — gating on either fix would silently remove a working feature. Without the built-in fix, only the telemetry agent is touched and the skipped daemons are named on screen. klipper, nexusp, quintusp and vectorp sit in an enforced refuse-list that carries both the S and CS name forms, since the init script names vary by firmware; quintusp in particular owns the power-loss GPIO. mDNS is a separate opt-in, because disabling it breaks .local resolution and Creality Print discovery. Stopping all five moved available memory by 13.2 MiB on a 200 MB machine; the larger win is the removed background IO.

Both directions are written to fail safely. Every rename is guarded, because helper.sh sets set -e process-wide and functions.sh run() invokes the action as a bare $1 — an unguarded mv would kill the whole helper after the daemon had already been stopped. Neither direction overwrites an existing file, so a firmware update that recreates an init script cannot be clobbered by a later restore. The print-state check consults both port 7126 and 7125, treats only explicitly known-idle states as idle, and falls through to a confirmation prompt on anything it cannot parse. The completion message reports how many services actually changed rather than asserting success, and warns if a process is still running.

Worth knowing: /usr/apps/etc/init.d is on p8 and survives a factory reset, so a reset will not undo this — the restore option is the way back. With solusp gone, action:cancel no longer appears in klippy.log; that string comes from Creality's own cancel handler, and the native pause_resume/cancel path triggers the quick stop directly, so its absence is expected rather than a regression.

Test plan

• Rename mechanism applied by hand on a K1C 2025 (V1.0.0.22.20250711S), printer idle: all five renamed and stopped, CS??* reduced to exactly the four survivors, klippy stayed ready, Fluidd/Moonraker/camera/nexusp still serving, ports 5353 and 8000 closed.
• Reboot-verified on that unit: all five stayed down across a real boot, and ports 5353 and 8000 did not return.
• Touchscreen Stop confirmed working with solusp disabled — cancel arrives as pause_resume/cancel from quintusp; print returned to standby, klippy stayed ready.
• Disable/restore logic exercised against a simulated init.d (not hardware): USB-only install leaves the WebRTC tier running, built-in fix unlocks it, a failed rename reports and does not kill the helper, restore preserves a firmware-recreated script, the print guard returns idle/printing/undetermined correctly including a decoy "state" key, and the refuse-list blocks both S and CS forms.
• The helper menu flow itself (disable → restore → disable via the new Customize entries) exercised on-printer.

Not addressed

Removing the Built-in Camera Fix while these services are disabled leaves the printer with no camera server at all, and usb_camera.sh has no awareness of this option to warn about it. Fixing that means touching the camera feature's own removal paths, so it is left for a follow-up rather than widened into this PR.

No firmware allowlist, unlike Restore Input Shapers: every rename is existence-guarded, so a firmware with a different daemon set is a no-op rather than a breakage. With solusp disabled, quintusp still requests AI detection at print start and waits out a ~10 s timeout before giving up — harmless, but it delays print start by a few seconds. Setting ai_detect.enable: 0 in system_config.json removes it; that isn't automated here because "enable" appears in four blocks of that file and a careless edit would also flip powerloss_continue.