fix(propagation): dedup PropagationGridWorker enqueues

FreshnessMonitor's comment claimed 'Oban unique constraint on
PropagationGridWorker prevents duplicates' — but the worker declared
no unique: option. During a long outage the monitor's 5-min stale-check
tick would pile up identical jobs (a 2h outage = 24 stacked jobs, each
doing the full f00-f18 HRRR chain).

Put the dedup on the worker (vs. the monitor's insert call) so anything
enqueuing it benefits — FreshnessMonitor, the hourly cron, or a manual
mix propagation_grid run.

unique: [period: 3600, states: [:available, :scheduled, :executing,
:retryable]] aligns with the hourly cron cadence. The seed job args
(%{}) collapse with themselves while chain-step jobs keep distinct
run_time/forecast_hour and remain enqueuable.

lib/microwaveprop/propagation/freshness_monitor.ex

@@ -54,9 +54,11 @@ defmodule Microwaveprop.Propagation.FreshnessMonitor do
   end
 
   defp enqueue_if_not_queued do
-    # Oban unique constraint on PropagationGridWorker prevents duplicates —
-    # if a job is already available/scheduled/executing/retryable within the
-    # last hour, insert is a no-op.
+    # PropagationGridWorker declares `unique:` over a 1-hour window on
+    # the seed args (`%{}`), so repeated 5-minute ticks during a long
+    # outage collapse into a single seed job — not one stacked chain
+    # per tick. `Oban.insert` returns `{:ok, job}` either way; the
+    # `conflict?` field on the returned job distinguishes the two.
     Oban.insert(PropagationGridWorker.new(%{}))
   end
 end

lib/microwaveprop/workers/propagation_grid_worker.ex

@@ -24,7 +24,16 @@ defmodule Microwaveprop.Workers.PropagationGridWorker do
     # (e.g., a rolling deploy that kills a mid-flight chain step)
     # don't exhaust the chain's retry budget and discard the whole
     # run. Legitimate scoring errors still give up after 5 attempts.
-    max_attempts: 5
+    max_attempts: 5,
+    # Deduplicate identical jobs across a 1-hour window. Uniqueness
+    # is over the full args set, so the seed (`%{}`) collapses with
+    # itself — FreshnessMonitor's 5-minute ticks during a long outage
+    # no longer stack 24 redundant f00-f18 chains — while chain steps
+    # with distinct `run_time` + `forecast_hour` args remain distinct
+    # from each other and from the seed. The period is the same as
+    # the hourly cron interval so successive hourly cron fires move
+    # past the window naturally.
+    unique: [period: 3600, states: [:available, :scheduled, :executing, :retryable]]
 
   alias Microwaveprop.Commercial
   alias Microwaveprop.Propagation

test/microwaveprop/propagation/freshness_monitor_test.exs

@@ -95,11 +95,12 @@ defmodule Microwaveprop.Propagation.FreshnessMonitorTest do
       end)
     end
 
-    test "repeated ticks on a persistently stale grid each enqueue another job" do
-      # Characterization note: the module comment claims an Oban unique
-      # constraint on PropagationGridWorker suppresses duplicates, but
-      # the worker currently declares no `unique:` option. Each tick
-      # therefore inserts a fresh job.
+    test "repeated ticks on a persistently stale grid are deduplicated into a single job" do
+      # PropagationGridWorker declares `unique:` on the seed args (`%{}`),
+      # so a stale-check tick that runs while an earlier seed is still
+      # available/scheduled/executing/retryable is a no-op. Without this,
+      # a long outage stacks one expensive HRRR f00-f18 chain per 5-min
+      # tick (e.g. a 2-hour outage = 24 redundant seed jobs).
       ScoresFile.write!(@band_mhz, minutes_ago(180), @sample_scores)
 
       Oban.Testing.with_testing_mode(:manual, fn ->
@@ -107,7 +108,7 @@ defmodule Microwaveprop.Propagation.FreshnessMonitorTest do
         FreshnessMonitor.handle_info(:check, %{})
         FreshnessMonitor.handle_info(:check, %{})
 
-        assert 3 == length(all_enqueued(worker: PropagationGridWorker))
+        assert 1 == length(all_enqueued(worker: PropagationGridWorker))
       end)
     end