Keep HTTP responsive during confluence replay.

Boot replay was a tight CPU loop that wedged the Unraid backend, so login hit the 8s client timeout. Yield every few as-of dates, shorten the boot ledger fill, and make /health a cheap liveness check.

app/server/src/confluence/confluenceEngine.ts

@@ -301,7 +301,12 @@ export async function runConfluenceReplay(
     const weeklyFull = await provider.resolveAsOf(symbol, '1wk', '9999-12-31');
     const benchFull = await provider.resolveAsOf(BENCHMARK_SYMBOL, '1d', '9999-12-31');
 
+    let asOfN = 0;
     for (const asOf of slice) {
+      asOfN += 1;
+      if (asOfN % 4 === 0) {
+        await new Promise<void>((r) => setImmediate(r));
+      }
       const dailyCut = daily.candles.filter((c) => (c.ts ?? '').slice(0, 10) <= asOf);
       const weeklyCut = weeklyFull.candles.filter((c) => (c.ts ?? '').slice(0, 10) <= asOf);
       const benchCut = benchFull.candles.filter((c) => (c.ts ?? '').slice(0, 10) <= asOf);
@@ -391,6 +396,7 @@ export async function fillLearningLedger(
       complete: batch.complete,
     };
     if (batch.evaluationsStored === 0) break;
+    await new Promise<void>((r) => setImmediate(r));
   }
   return { ...last, ledger: await learningLedgerStatus(db, cache) };
 }