A failed plan-force state read journals blocked rows that hold the plans for a day #4769

The plan-force bot's state read has a 5-second deadline. Any fault returns a null state with a reason, and each candidate (up to 10) gets a blocked row. Those rows feed last_journaled, so the 24-hour cooldown then hides the real verdict for a day.

The row itself is honest. The cooldown ranks above the state blockers on purpose, and a test pins that order.

Where (as of 30cbbcf)

• The deadline: Darling/PerformanceMonitor.Darling.Service/ServiceCommandDeadlines.cs:157.
• The blocked rows: Darling/PerformanceMonitor.Darling.Service/PlanForceBot.cs:153 and :187-212.
• last_journaled: Darling/PerformanceMonitor.Darling.Service/PgPlanForceActionStore.cs:236-243 and :317-324.
• The order: PerformanceMonitor.Analysis/ForcePlanBotPolicy.cs:391-399, pinned by TheCooldown_StillOutranksTheStateBlockers (Darling/Darling.Tests/PlanForceBotTests.cs:479).

Fix

Keep that order, but give state_unavailable its own short window of about 1 hour instead of the 24-hour cooldown. If it is dropped from the cooldown altogether, a long outage floods the journal.

Test that must fail first

• After a state-read failure, the candidate is judged again after the short window, not after 24 hours.
• Repeated failures inside the window journal once.