feat(discovery): acquire recommendations that keep their names

Acquisition ran yt-dlp without --embed-metadata, so every download
arrived untagged. The scanner then stored the video id as the title and
"Unknown Artist" as the artist, the vetted-candidate tag check rejected
the mismatch, and all 18 acquired tracks were hidden and retired.

- Pass --embed-metadata so downloads carry real tags.
- Let a scan take fallback title/artist from the candidate, for sources
  that still ship untagged files.
- Install Deno alongside yt-dlp: YouTube guards some formats with a JS
  challenge yt-dlp must execute, and no other runtime is enabled.
- Dedupe candidates by artist and title. The (source, external_id) key
  misses the same song reaching us under two Deezer release ids.

Also carries the in-flight discovery work this builds on: the
Recommendations page replacing Discover, the discovery source service,
and the acquisition spec tests.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

backend/src/app.ts

@@ -18,6 +18,11 @@ import vibeSessionsRoutes from './routes/vibe-sessions.routes.js';
 import discoveryRoutes from './routes/discovery.routes.js';
 import imagesRoutes from './routes/images.routes.js';
 import { VibeSessionCoordinator } from './services/vibe-session-coordinator.service.js';
+import { DiscoveryService } from './services/discovery.service.js';
+
+// Single-listener deployment: the same placeholder the vibe-session routes use
+// when no x-user-id header is supplied.
+const DEFAULT_USER_ID = '00000000-0000-0000-0000-000000000000';
 
 export interface AppConfig {
   port: number;
@@ -91,6 +96,34 @@ export async function buildApp(config: AppConfig) {
   dbService.decayBeliefs().catch(() => {});
   dbService.deriveForgottenProfile().catch(() => {});
 
+  // Close the discovery loop without an operator in it. The worker's cron jobs
+  // generate candidates; this timer is what decides which of them earn a
+  // probation slot and enqueues the download. Evaluation is cheap SQL against a
+  // capped batch, and every acquisition gate (relevance, novelty tolerance,
+  // backlog, per-artist diversity) lives inside evalCandidates.
+  const discoveryService = new DiscoveryService(dbService);
+  const DISCOVERY_EVAL_INTERVAL_MS = 30 * 60 * 1000;
+  const runDiscoveryEval = async () => {
+    const results = await discoveryService.evalCandidates(DEFAULT_USER_ID);
+    let enqueued = 0;
+    for (const result of results) {
+      if (!result.shouldAcquire) continue;
+      try {
+        await jobService.enqueueDiscoveryAcquisition(result.candidateId);
+        enqueued++;
+      } catch (err) {
+        const reason = err instanceof Error ? err.message : 'failed to enqueue acquisition';
+        await discoveryService.markEnqueueFailed(result.candidateId, reason);
+      }
+    }
+    if (results.length > 0) {
+      console.log(`[Discovery] Evaluated ${results.length} candidates, enqueued ${enqueued}`);
+    }
+  };
+  const discoveryEvalTimer = setInterval(() => {
+    runDiscoveryEval().catch((e) => console.error('[Discovery] eval failed:', e));
+  }, DISCOVERY_EVAL_INTERVAL_MS);
+
   // Ensure the Typesense 'tracks' collection schema exists on boot so that
   // the first search request doesn't hit a 404.
   await searchService.ensureCollection();
@@ -176,6 +209,7 @@ export async function buildApp(config: AppConfig) {
       clearInterval(fusionTimer);
       clearInterval(decayTimer);
       clearInterval(forgottenTimer);
+      clearInterval(discoveryEvalTimer);
     } catch (err) {
       fastify.log.error(err);
     }