fix(video): a cue aimed at an output the node doesn't drive stays dark (869fgq3xv)

[ibiltari]

ClickUp: 869fgq3xv · Plan: cuems-RELATIONS/Plans/2026-10-13-engine-video-cue-unrouted-output.md (reviewed by Sonnet; decisions D1–D6 by Ion)

Problem

A video cue pointing at an output this node does not drive was loaded anyway. That covers an output dropped at node start (DISPLAY_MISMATCH, no region in display.conf), one with enabled=false, or a custom_<n> with no VideoCueOutput. With no position/scale, the videocomposer letterboxed the layer onto the whole canvas, so the content appeared on the screens that do work. The only trace was a "position/scale NOT applied" line.

Change (node engine only, log only)

• PlayerHandler.partition_cue_outputs splits a cue's outputs into routed and unrouted. The arm and the load-time preflight both use it, so they cannot disagree. Any resolver exception counts as unrouted; nothing escapes arm_videoCue any more.
• arm_videoCue loads routed outputs only.
  • The shared-decoder driver is the first routed layer.
  • Layer ids keep the output's index (<cue>_<i>).
  • cue._layer_outputs records layer → output.
  • A cue with every output dark still arms, so it doesn't hold the GO gate and doesn't trigger the GO fallback re-arm.
  • The error is logged once per load, then at DEBUG.
• run_videoCue stamps _start_mtc/_end_mtc before its layer check, so a dark cue keeps its place on the timeline. Before this, zero timecodes made a go_at_end follow fire at start + postwait cluster-wide, and made an infinite loop re-base in a tight loop. Layers now map to outputs by id, not by position.
• read_display_conf drops a section whose enabled= is anything but true/1, even when it keeps its canvas_region. It parses the way DisplayConfigurationManager::loadFromFile does: no trimming, last line wins. Repeated keys no longer raise.
• NodeEngine.check_unrouted_video_outputs logs UNROUTED_VIDEO_OUTPUT per dark cue at each load, before anything arms. It runs in _load_project_inner, not ready_project, because ready_project is also the mid-show deploy command.

Behaviour change on upgrade: a box whose journal shows DISPLAY_MISMATCH, or that has an enabled=false section with a region, goes from mis-placed video to dark for the cues aimed there. This reverses 869f8hfra's "no output match still plays". A resolved output whose placement send fails still plays.

Out of scope:

• The engine still reads display.conf only at node start. After a hand edit, restart the videocomposer and the node engine.
• Nothing is reported to the UI; that waits for the display configuration manager.
• No message between processes changes.

Tests

• New: tests/test_video_unrouted_output.py and more cases in tests/test_display_conf.py.
• The 869f8hfra placement tests now inject their failure at the apply step.
• Red on rc_1 (12 failed, 26 errors). Green on test2: full suite 1334 passed.
• black / isort / flake8 clean.

Not yet verified on a rig. §6 of the plan, on the test cluster, is still owed.