fix(quotas): unifier le décompte IA (BYOK, rollback) et combler les fuites

Centralise la réserve via ai-quota, corrige admin unavailable (-1), brancher les routes sans quota et le host-pays brainstorm, avec usage-meter élargi, noms de clusters, MCP et ajustements dashboard/insights.

Co-authored-by: Cursor <cursoragent@cursor.com>

memento-note/lib/entitlements.ts

@@ -190,6 +190,46 @@ local newCount = tonumber(redis.call('GET', KEYS[1]))
 return newCount
 `;
 
+const RELEASE_LUA = `
+local current = tonumber(redis.call('GET', KEYS[1]) or '0')
+if current <= 0 then return 0 end
+redis.call('DECR', KEYS[1])
+return tonumber(redis.call('GET', KEYS[1]) or '0')
+`;
+
+export async function releaseUsage(userId: string, feature: string): Promise<void> {
+  if (!isValidFeature(feature)) return;
+
+  const tier = await getEffectiveTier(userId);
+  const limit = await getLimitAsync(tier, feature);
+  if (limit === undefined || limit === Infinity) return;
+
+  try {
+    const key = getRedisKey(userId, feature);
+    const raw = await redis.eval(RELEASE_LUA, 1, key);
+    const newCount = parseReserveResult(raw);
+    if (Number.isFinite(newCount) && newCount >= 0) {
+      await mirrorUsageCountToDatabase(userId, feature, newCount);
+    }
+  } catch (err) {
+    console.error('[entitlements] Redis release failed, trying DB fallback:', err);
+    try {
+      const { periodStart } = getPeriodDates();
+      const existing = await prisma.usageLog.findUnique({
+        where: { userId_feature_periodStart: { userId, feature, periodStart } },
+      });
+      const current = existing?.requestsCount ?? 0;
+      if (current <= 0) return;
+      await prisma.usageLog.update({
+        where: { userId_feature_periodStart: { userId, feature, periodStart } },
+        data: { requestsCount: current - 1, syncedAt: new Date() },
+      });
+    } catch (dbErr) {
+      console.error('[entitlements] DB release fallback failed:', dbErr);
+    }
+  }
+}
+
 const RESERVE_LUA = `
 local limit = tonumber(ARGV[1])
 local ttl = tonumber(ARGV[2])
@@ -472,6 +512,9 @@ export async function getUserQuotas(
     return result;
   } catch (err) {
     console.error('[entitlements] getUserQuotas Redis error:', err);
+    if (shouldFailClosedOnRedisError()) {
+      throw new QuotaServiceUnavailableError();
+    }
     const result: Record<string, { remaining: number; limit: number; used: number }> = {};
     for (const feature of features) {
       const limit = (await getLimitAsync(tier, feature)) ?? 0;