Verify extraction provenance and prune semantic staging before inference

crates/searchi-service/src/media.rs

@@ -42,7 +42,7 @@ impl HybridSearch {
                 limits.max_file_bytes
             ));
         }
-        self.rebuild_generation(fingerprint, |service, generation, work| {
+        self.rebuild_generation(fingerprint, Some(limits), |service, generation, work| {
             service.build(generation, work)?;
             let mut after = 0;
             let mut media_count = 0usize;