The code indexer currently caches cloned repositories on local disk with a three-tier resolution strategy (cache hit, incremental update, full download fallback). At scale this is ineffective: - With 36 indexer pods, the cache hit rate is effectively zero since consecutive indexing tasks for the same repo are unlikely to land on the same pod - No eviction policy means disk usage grows unbounded, risking the 10GB pod limit - The incremental update path adds complexity with no practical benefit This MR removes the disk cache entirely. The indexer now downloads the full repository archive from the Rails internal API on every indexing task and deletes the files from disk immediately after indexing completes.