emrg: GUI task primary/secondary list — drop last_cycle_summary from primary row + add reason column to run detail

[argszero]

Fixes host rant 2026-08-19T18:25:14 (GUI 任务管理·一级/二级列表).

Problem

• The task management primary list showed last_cycle_summary ("what was done") inline, cluttering the row.
• The click-to-expand secondary list (run records) lacked the reason (原因) for the throttle/idle judgment.

Changes

• emrg/gui/renderer/js/dialogs.js: primary task row no longer renders last_cycle_summary; accordion head/rows gain a 4th Reason column (app.taskRunsColReason).
• emrg/server/scheduler.py + emrg/protocol.py: EvolutionLog gains reason; the vibe-check reason field (natural-language judgment explanation) is now persisted and surfaced in status()recent_runs.
• emrg/gui/renderer/js/i18n.js: new app.taskRunsColReason key (zh/en); orphaned app.taskLastRunSummary removed (no longer referenced).
• emrg/gui/renderer/css/components.css: run-detail grid widened to 4 columns (72px / 1fr / 92px / 1fr) + ellipsized .task-run-reason.
• Tests: scheduler recent_runs reason assertions; GUI accordion test asserts reason column + primary list no longer shows the summary.

Verification

• Python suite: 976 passed + 1 skipped (977 collected)
• GUI i18n + renderer smoke: 147/147; daemon_client: 45/45
• Import + CLI checks pass

Note: GUI integration tests are skipped locally while the daemon holds the fixed port (single-instance admission); they run in CI's isolated environment.

[argszero]

✅ LGTM — cycle

Verified:

• diff is 7 files / +45 −13, scoped to the rant's requirement (primary row drops last_cycle_summary; accordion gains a Reason column fed by a newly persisted vibe-check reason field).
• Scheduler status() recent_runs now carries reason; EvolutionLog dataclass extended; no daemon API shape changes elsewhere.
• i18n: new app.taskRunsColReason (zh/en), orphaned app.taskLastRunSummary removed — orphan/missing-key guard passes.
• Local verification: Python suite 976 passed + 1 skipped (977 collected, matches Agent.md); GUI i18n + renderer smoke 147/147; daemon_client 45/45; import + CLI OK.
• CI: test job PASS. (test-windows still running; integration tests need an isolated daemon so they are skipped locally by design while the daemon holds the fixed port.)

[argszero]

✅ LGTM — cycle (2nd)

Re-verified in this cycle:

• Head unchanged (6f4b09b, 7 files / +45 −13); MERGEABLE.
• Change is scoped and consistent: primary row drops last_cycle_summary; accordion gains Reason column fed by the newly persisted vibe-check reason (EvolutionLog + status().recent_runs); i18n orphan app.taskLastRunSummary removed; CSS grid 4 cols.
• Local: Python suite 976 passed + 1 skipped (977 = Agent.md); GUI i18n+smoke 147/147; daemon_client 45/45; import + CLI OK.
• CI: original run's test job PASS; the Windows job was queue-stuck for 35+ min (same start time as the 1m25s linux job) — cancelled and re-triggered via workflow_dispatch (run 32245064133), now in progress.

[argszero]

✅ LGTM — cycle (3rd)

Final verification:

• Head b369899 (empty re-check commit on top of 6f4b09b, no code change — supersedes the cancelled Windows check from the old run); mergeStateStatus CLEAN, MERGEABLE.
• Fresh pull_request run 32245663591: test PASS (1m34s) + test-windows PASS (58s).
• Local verification from cycles 930/931 stands: Python 976+1=977, GUI i18n+smoke 147/147, daemon_client 45/45, import + CLI OK.
• 3 consecutive LGTMs (cycles 930/931/932), no ❌ between → merging.