Add serialization API for set of rows (#7883)

Summary:
Top level rows are more efficiently serialized in row sets rather than arrays or ranges. Arrays of ranges are still useful for repeated nested content.  The row set path can uses SIMD to gather nulls and extract idices of non-null values for serialization.

A Scratch objett is added to signatures to pass reusable scratch memory also for top level calls to range serializing serializatin functions. This can remove malloc use for temporary vectors.

The API is tested standalone but is not connected to running code, so this diff does not affect running systens.

Pull Request resolved: #7883

Reviewed By: Yuhta

Differential Revision: D51858722

Pulled By: oerling

fbshipit-source-id: 5d48a3f57df30f56818afdbcee39fc8602a34984

velox/serializers/CompactRowSerializer.cpp

@@ -22,7 +22,8 @@ namespace facebook::velox::serializer {
 void CompactRowVectorSerde::estimateSerializedSize(
     VectorPtr /* vector */,
     const folly::Range<const IndexRange*>& /* ranges */,
-    vector_size_t** /* sizes */) {
+    vector_size_t** /* sizes */,
+    Scratch& /*scratch*/) {
   VELOX_UNSUPPORTED();
 }
 
@@ -36,7 +37,8 @@ class CompactRowVectorSerializer : public VectorSerializer {
 
   void append(
       const RowVectorPtr& vector,
-      const folly::Range<const IndexRange*>& ranges) override {
+      const folly::Range<const IndexRange*>& ranges,
+      Scratch& scratch) override {
     size_t totalSize = 0;
     row::CompactRow row(vector);
     if (auto fixedRowSize =

velox/serializers/CompactRowSerializer.h

@@ -28,7 +28,8 @@ class CompactRowVectorSerde : public VectorSerde {
   void estimateSerializedSize(
       VectorPtr vector,
       const folly::Range<const IndexRange*>& ranges,
-      vector_size_t** sizes) override;
+      vector_size_t** sizes,
+      Scratch& scratch) override;
 
   // This method is not used in production code. It is only used to
   // support round-trip tests for deserialization.