Stop per-tick log flooding when referenced scenes are absent

[lvkz]

When the active scene collection doesn't contain the scenes a deck page references — e.g. switching OBS to a lightweight "idle" collection while the page targets the full one — two per-tick status polls flood the log for as long as that collection stays active:

1. SceneItem actions log Cannot find the scene item! unconditionally on every tick. A page with 58 SceneItem actions writes ~3,500 journal lines per minute.
2. Filter actions crash on every tick: get_source_filter returns an empty dict (not None) when the scene/filter is missing, so the status is None guard passes and status["filterEnabled"] raises a KeyError through the rpyc netref — loguru then dumps a ~65-line traceback once per second.

On the affected machine this produced ~7,500 journal lines/minute, a 1 GB journald and 320 MB of app logs over five days.

Fixes

• OBSController.get_scene_item_enabled: track already-reported (scene, item) pairs and warn once per pair, naming the pair so the message is actionable.
• FilterBase.show_current_filter_status: treat an empty/keyless status as the existing UNKNOWN path instead of indexing it.

Testing

Applied to a live installation (StreamController 1.4.11-beta, OBS 32 with the deck page above, OBS parked in the idle collection):

last 60 s	before	after
scene-item warnings	~3,500	0 (19 one-shot warnings at startup, one per unique pair)
filter tracebacks	~60	0
total plugin journal lines	~7,500	25

Button behavior is unchanged — both patches touch only the read/status path; press handlers (set_scene_item_enabled, filter toggles) are untouched, and state display still resolves normally once the right collection is active.