perf: optimize recommendation cache misses

[Priveetee]

Summary

• Return recommendation cache misses from fast cache-only paths while full builds warm asynchronously.
• Add stale recommendation cache, longer feed/pool TTLs, and proactive warmup after auth/subscription changes.
• Reduce background related-stream workload for large accounts.

Tests

• JAVA_HOME=/usr/lib/jvm/java-21-openjdk ./gradlew test
• dev CI, Coverage, and Docker workflows passed for 35d220e and 9982f78
• Beta Docker image tested locally on /recommendations/home and /recommendations/shorts