Remove per-chunk flushing to speedup AR write speed.

This gives a huge performance difference for small group size in the pygrain
use case.

PiperOrigin-RevId: 772929139

Author: dryman

cpp/sequenced_chunk_writer.cc

@@ -126,12 +126,6 @@ void SequencedChunkWriterBase::TrySubmitFirstFutureChunk(
       return;
     }
   }
-  if (!chunk_writer->Flush(riegeli::FlushType::kFromObject)) {
-    Fail(riegeli::Annotate(
-        chunk_writer->status(),
-        absl::StrFormat("Could not flush chunk: %d", submitted_chunks_)));
-    return;
-  }
   if (callback_) {
     (*callback_)(submitted_chunks_, chunk_offset, decoded_data_size,
                  num_records);
@@ -148,9 +142,6 @@ void SequencedChunkWriterBase::Initialize() {
     Fail(riegeli::Annotate(chunk_writer->status(),
                            "Failed to create the file header"));
   }
-  if (!chunk_writer->Flush(riegeli::FlushType::kFromObject)) {
-    Fail(riegeli::Annotate(chunk_writer->status(), "Could not flush"));
-  }
 }
 
 void SequencedChunkWriterBase::Done() {