MC-4397

Implement the operator-noise root-cause fixes surfaced by MC-4333 and Elmar's question about pending tickets not moving. Context: - MC-4333 fixed the immediate ARG_MAX crash in scripts/luci_operator_tuner.py and refreshed reports/operator-tune-2026-05-28.md. - The refreshed report identifies high-confidence fixes that stop technical/operator noise from becoming misleading pending work. Scope: 1. In luci_operator.py, classify workspace graphify-out/ output as known generated dirty noise (converted/ and cache/), so graphify churn does not repeatedly create dirty-repo/operator tickets. 2. Raise ACTIVE_LANE_STALE_HOURS_TRIGGER from 2 to 4 so legitimate multi-hour worker lanes are not escalated as stuck every cycle. 3. Reduce duplicate recent_task_failure logging across overlapping operator runs: dedupe by task_run id or narrow the query window enough that the same failed task_run is not logged ~4x. 4. Preserve the MC-4381 rule: routine operator-tune reports go to orchestrator_inbox/Luci, not Elmar-facing tickets/Telegram. Acceptance: - Code committed and pushed to workspace master. - Relevant narrow tests / py_compile pass. - Run a safe dry-run or unit-level verification showing graphify-out is known-generated, threshold is 4h, and recent task failure logging no longer repeats the same task_run every cycle. - Board check after landing shows no runnable technical/control-plane ticket assigned to Elmar.