Send smaller first batch (5 features) for faster first paint

Subsequent batches use the normal batch_size (default 50). This
reduces server-side time-to-first-property by ~10x since only 5
features need to be serialized before the first yield.

api/app.py

@@ -53,6 +53,7 @@ configure_logging("api")
 logger = logging.getLogger(__name__)
 
 DEFAULT_BATCH_SIZE = 50
+FIRST_BATCH_SIZE = 5
 _rate_limit_config = RateLimitConfig.from_env()
 
 
@@ -288,6 +289,7 @@ async def _stream_from_cache(
     }) + "\n"
 
     count = 0
+    is_first_batch = True
     for feature_batch in get_cached_features(query_parameters, batch_size=batch_size):
         # Apply decision filtering
         if decision_filter != "everything":
@@ -302,9 +304,27 @@ async def _stream_from_cache(
             ]
         if limit and count + len(feature_batch) > limit:
             feature_batch = feature_batch[:limit - count]
-        count += len(feature_batch)
-        if feature_batch:
-            yield json.dumps({"type": "batch", "features": feature_batch}) + "\n"
+
+        # Split first batch into smaller primer batch and remainder
+        if is_first_batch and len(feature_batch) > FIRST_BATCH_SIZE:
+            # Yield primer batch
+            first_features = feature_batch[:FIRST_BATCH_SIZE]
+            count += len(first_features)
+            yield json.dumps({"type": "batch", "features": first_features}) + "\n"
+
+            # Yield remainder
+            remaining_features = feature_batch[FIRST_BATCH_SIZE:]
+            count += len(remaining_features)
+            if remaining_features:
+                yield json.dumps({"type": "batch", "features": remaining_features}) + "\n"
+            is_first_batch = False
+        else:
+            # Normal batch yielding
+            count += len(feature_batch)
+            if feature_batch:
+                yield json.dumps({"type": "batch", "features": feature_batch}) + "\n"
+            is_first_batch = False
+
         if limit and count >= limit:
             break
 
@@ -341,6 +361,7 @@ async def _stream_from_db(
     try:
         count = 0
         batch: list[dict] = []
+        current_batch_target = FIRST_BATCH_SIZE  # Start with smaller first batch
         for row in repository.stream_listings_optimized(
             query_parameters, limit=limit, page_size=batch_size
         ):
@@ -359,11 +380,13 @@ async def _stream_from_db(
             batch.append(feature)
             count += 1
 
-            if len(batch) >= batch_size:
+            if len(batch) >= current_batch_target:
                 if staging_key:
                     cache_features_batch_staged(staging_key, batch)
                 yield json.dumps({"type": "batch", "features": batch}) + "\n"
                 batch = []
+                # After first batch, use normal batch size
+                current_batch_target = batch_size
 
         if batch:
             if staging_key: