feat(status): report the printer's own state, and the real print outcome

[impuls42]

Pairs with sengine-cloud/labelfab#9, which publishes the device truth this consumes.

The problem

The Machines page could say connected about a printer with no tape in it. Status collapsed to three cases, so every fault the agent published looked healthy:

if status is None:                          -> UNKNOWN
elif status.get("state") == "disconnected": -> DISCONNECTED
else:                                       -> CONNECTED     # including state == "error"

And print_labels set PRINTING and returned. JobResult was published by the agent on <prefix>/<printer_id>/results and consumed by nobody, so InvenTree never learned whether a job succeeded.

Status mapping

Uses the codes InvenTree already defines (CONNECTED 100, UNKNOWN 101, PRINTING 110, NO_MEDIA 301, DISCONNECTED 400, ERROR 500):

Printer reports	Machines status
nothing retained	UNKNOWN — agent has never published
idle, media ok	CONNECTED — with fw, battery, voltage, serial
media bit clear (0x06)	NO_MEDIA
any other fault	ERROR, carrying the reason
agent will / shutdown	DISCONNECTED

Two deliberate distinctions:

• A media fault is not a generic error. NO_MEDIA and ERROR render differently and mean different things to whoever is looking: "go load tape" vs "go investigate".
• Unreported media is not healthy media. media_ok is a tri-state; a printer that has said nothing shows media unreported. Treating silence as health is exactly how the old mapping produced a confident, wrong answer.

Battery percentage and voltage are both shown because the percentage pins at 100% whenever the unit is on charge — measured on the bench, 100% for the entire time voltage was climbing 4.16→4.17 V. On its own it cannot answer whether the printer will survive a long strip.

Print outcomes

AWAIT_RESULT_S waits for the job's JobResult, so the UI reports what happened rather than that something was sent.

• Subscribes before publishing. The results topic is not retained, so subscribing afterwards races the agent and loses a fast print's result entirely.
• Filters on job_id. One printer serves every InvenTree user; another click's result must not be mistaken for yours.
• A timeout is an ordinary outcome. Strip mode holds labels until the coalescer flushes (30 s by default), which is longer than any wait worth blocking a Django-Q worker on. It stays PRINTING, and the next refresh reads the truth off the retained topic.
• partial_tape_consumed is surfaced in the text, since a blind reprint after one double-prints.

It defaults to 0 (off). The wait holds a Django-Q worker, and in strip mode the agent buffers until its coalescer flushes (30 s), so it would usually time out and leave the status to the retained topic regardless. Worth enabling on die-cut/discrete media, where a result lands in a few seconds. The help text states the worker-blocking cost explicitly.

Structure

The mapping lives in a new status.py with no InvenTree imports, so it is unit-testable here — InvenTree is intentionally not a dependency of this package. The driver stays the thin adapter it claims to be, translating a returned member name to the enum.

test_every_mapping_names_a_real_status_member pins every name against the real LabelPrinterStatus members, because a typo would otherwise surface as an AttributeError inside a Django worker rather than at test time.

34 tests pass, ruff clean.

PAPER_JAM (302) is deliberately unreachable: labelfab's contract state is only idle/printing/disconnected/error, and HARDWARE-NOTES records that a jam has no known encoding on this printer, so there is nothing to map it from.