[opt](memory) implement freeing memory column by column for partial update (apache#58275)

### What problem does this PR solve?

#### Background
Now, pages of column writer will be released after write all memtable
block in the scenario of partial updates, the process is as follows:
1. create all column_writer
2. append all column_writer (significant amount of memory expansion)
3. finish and free memory.

#### Problem 
This implement will cause a significant amount of memory expansion when
flushing memtable(The larger the number of columns, the more obvious it
is), and make load could easily cancelled by memory manager when table
column too wide(such as 5000 column).

#### Solution
This pr implement freeing memory column by column for partial update to
solve the problem, the process is as follows:
1. create all column_writer
2. append column_writer and free memory column by column.

#### Test results
Test using machines with specifications of 16c and 64G with 5000 column
table:

- Before this pr, `VerticalSegmentWriter::_create_column_writer`
occupies nearly 20% of the memory.
- After fix, ` VerticalSegmentWriter::_create_column_writer` hardly
occupies memory.

Author: sollhui

be/src/olap/rowset/segment_v2/vertical_segment_writer.cpp

@@ -447,6 +447,19 @@ Status VerticalSegmentWriter::_probe_key_for_mow(
     return Status::OK();
 }
 
+Status VerticalSegmentWriter::_finalize_column_writer_and_update_meta(size_t cid) {
+    RETURN_IF_ERROR(_column_writers[cid]->finish());
+    RETURN_IF_ERROR(_column_writers[cid]->write_data());
+
+    auto* column_meta = _column_writers[cid]->get_column_meta();
+    column_meta->set_compressed_data_bytes(
+            _column_writers[cid]->get_total_compressed_data_pages_bytes());
+    column_meta->set_uncompressed_data_bytes(
+            _column_writers[cid]->get_total_uncompressed_data_pages_bytes());
+    column_meta->set_raw_data_bytes(_column_writers[cid]->get_raw_data_bytes());
+    return Status::OK();
+}
+
 Status VerticalSegmentWriter::_partial_update_preconditions_check(size_t row_pos,
                                                                   bool is_flexible_update) {
     if (!_is_mow()) {
@@ -536,6 +549,7 @@ Status VerticalSegmentWriter::_append_block_with_partial_content(RowsInBlock& da
         }
         RETURN_IF_ERROR(_column_writers[cid]->append(column->get_nullmap(), column->get_data(),
                                                      data.num_rows));
+        RETURN_IF_ERROR(_finalize_column_writer_and_update_meta(cid));
     }
 
     bool has_default_or_nullable = false;
@@ -625,6 +639,7 @@ Status VerticalSegmentWriter::_append_block_with_partial_content(RowsInBlock& da
         }
         RETURN_IF_ERROR(_column_writers[cid]->append(column->get_nullmap(), column->get_data(),
                                                      data.num_rows));
+        RETURN_IF_ERROR(_finalize_column_writer_and_update_meta(cid));
     }
 
     _num_rows_updated += stats.num_rows_updated;
@@ -722,6 +737,7 @@ Status VerticalSegmentWriter::_append_block_with_flexible_partial_content(
         RETURN_IF_ERROR(_column_writers[cid]->append(column->get_nullmap(), column->get_data(),
                                                      data.num_rows));
         DCHECK(_column_writers[cid]->get_next_rowid() == _num_rows_written + data.num_rows);
+        RETURN_IF_ERROR(_finalize_column_writer_and_update_meta(cid));
     }
 
     // 5. genreate read plan
@@ -767,6 +783,7 @@ Status VerticalSegmentWriter::_append_block_with_flexible_partial_content(
         RETURN_IF_ERROR(_column_writers[cid]->append(column->get_nullmap(), column->get_data(),
                                                      data.num_rows));
         DCHECK(_column_writers[cid]->get_next_rowid() == _num_rows_written + data.num_rows);
+        RETURN_IF_ERROR(_finalize_column_writer_and_update_meta(cid));
     }
 
     _num_rows_updated += stats.num_rows_updated;
@@ -928,17 +945,6 @@ Status VerticalSegmentWriter::write_batch() {
                 RETURN_IF_ERROR(_append_block_with_partial_content(data, full_block));
             }
         }
-        for (auto& column_writer : _column_writers) {
-            RETURN_IF_ERROR(column_writer->finish());
-            RETURN_IF_ERROR(column_writer->write_data());
-
-            auto* column_meta = column_writer->get_column_meta();
-            column_meta->set_compressed_data_bytes(
-                    column_writer->get_total_compressed_data_pages_bytes());
-            column_meta->set_uncompressed_data_bytes(
-                    column_writer->get_total_uncompressed_data_pages_bytes());
-            column_meta->set_raw_data_bytes(column_writer->get_raw_data_bytes());
-        }
         return Status::OK();
     }
     // Row column should be filled here when it's a directly write from memtable
@@ -988,15 +994,7 @@ Status VerticalSegmentWriter::write_batch() {
             return Status::Error<DISK_REACH_CAPACITY_LIMIT>("disk {} exceed capacity limit.",
                                                             _data_dir->path_hash());
         }
-        RETURN_IF_ERROR(_column_writers[cid]->finish());
-        RETURN_IF_ERROR(_column_writers[cid]->write_data());
-
-        auto* column_meta = _column_writers[cid]->get_column_meta();
-        column_meta->set_compressed_data_bytes(
-                _column_writers[cid]->get_total_compressed_data_pages_bytes());
-        column_meta->set_uncompressed_data_bytes(
-                _column_writers[cid]->get_total_uncompressed_data_pages_bytes());
-        column_meta->set_raw_data_bytes(_column_writers[cid]->get_raw_data_bytes());
+        RETURN_IF_ERROR(_finalize_column_writer_and_update_meta(cid));
     }
 
     for (auto& data : _batched_blocks) {

be/src/olap/rowset/segment_v2/vertical_segment_writer.h

@@ -198,6 +198,8 @@ class VerticalSegmentWriter {
             vectorized::IOlapColumnDataAccessor* seq_column, size_t num_rows, bool need_sort);
     Status _generate_short_key_index(std::vector<vectorized::IOlapColumnDataAccessor*>& key_columns,
                                      size_t num_rows, const std::vector<size_t>& short_key_pos);
+    Status _finalize_column_writer_and_update_meta(size_t cid);
+
     bool _is_mow();
     bool _is_mow_with_cluster_key();