fix(physical-plan): set column byte_size to 0 in FilterExec zero-row interval stats (#21999)

## Which issue does this PR close?
- Closes #.

## Rationale for this change
While reading code around column statistics I've realized that we send
column byte_size when there are no rows. In this case I think it is
better to set it to 0

## What changes are included in this PR?
Set `byte_size` to zero when there are no rows in filter column
statistics

## Are these changes tested?
Yes existing unit tests

## Are there any user-facing changes?
no api changes and minimal propagated statistics change

Author: buraksenn

datafusion/core/tests/physical_optimizer/partition_statistics.rs

@@ -381,8 +381,6 @@ mod test {
         let filter: Arc<dyn ExecutionPlan> =
             Arc::new(FilterExec::try_new(predicate, scan)?);
         let full_statistics = filter.partition_statistics(None)?;
-        // Filter preserves original total_rows and byte_size from input
-        // (4 total rows = 2 partitions * 2 rows each, byte_size = 4 * 4 = 16 bytes for int32)
         let expected_full_statistic = Statistics {
             num_rows: Precision::Inexact(0),
             total_byte_size: Precision::Inexact(0),
@@ -393,15 +391,15 @@ mod test {
                     min_value: Precision::Exact(ScalarValue::Int32(None)),
                     sum_value: Precision::Exact(ScalarValue::Int32(None)),
                     distinct_count: Precision::Exact(0),
-                    byte_size: Precision::Exact(16),
+                    byte_size: Precision::Exact(0),
                 },
                 ColumnStatistics {
                     null_count: Precision::Exact(0),
                     max_value: Precision::Exact(ScalarValue::Date32(None)),
                     min_value: Precision::Exact(ScalarValue::Date32(None)),
                     sum_value: Precision::Exact(ScalarValue::Date32(None)),
                     distinct_count: Precision::Exact(0),
-                    byte_size: Precision::Exact(16), // 4 rows * 4 bytes (Date32)
+                    byte_size: Precision::Exact(0),
                 },
             ],
         };
@@ -411,7 +409,6 @@ mod test {
             .map(|idx| filter.partition_statistics(Some(idx)))
             .collect::<Result<Vec<_>>>()?;
         assert_eq!(statistics.len(), 2);
-        // Per-partition stats: each partition has 2 rows, byte_size = 2 * 4 = 8
         let expected_partition_statistic = Statistics {
             num_rows: Precision::Inexact(0),
             total_byte_size: Precision::Inexact(0),
@@ -422,15 +419,15 @@ mod test {
                     min_value: Precision::Exact(ScalarValue::Int32(None)),
                     sum_value: Precision::Exact(ScalarValue::Int32(None)),
                     distinct_count: Precision::Exact(0),
-                    byte_size: Precision::Exact(8),
+                    byte_size: Precision::Exact(0),
                 },
                 ColumnStatistics {
                     null_count: Precision::Exact(0),
                     max_value: Precision::Exact(ScalarValue::Date32(None)),
                     min_value: Precision::Exact(ScalarValue::Date32(None)),
                     sum_value: Precision::Exact(ScalarValue::Date32(None)),
                     distinct_count: Precision::Exact(0),
-                    byte_size: Precision::Exact(8), // 2 rows * 4 bytes (Date32)
+                    byte_size: Precision::Exact(0),
                 },
             ],
         };

datafusion/physical-plan/src/filter.rs

@@ -903,7 +903,7 @@ fn collect_new_statistics(
                         min_value: Precision::Exact(typed_null.clone()),
                         sum_value: Precision::Exact(typed_null),
                         distinct_count: Precision::Exact(0),
-                        byte_size: input_column_stats[idx].byte_size,
+                        byte_size: Precision::Exact(0),
                     };
                 };
                 let (lower, upper) = interval.into_bounds();
@@ -1621,15 +1621,15 @@ mod tests {
                     sum_value: Precision::Exact(ScalarValue::Int32(None)),
                     distinct_count: Precision::Exact(0),
                     null_count: Precision::Exact(0),
-                    byte_size: Precision::Absent,
+                    byte_size: Precision::Exact(0),
                 },
                 ColumnStatistics {
                     min_value: Precision::Exact(ScalarValue::Int32(None)),
                     max_value: Precision::Exact(ScalarValue::Int32(None)),
                     sum_value: Precision::Exact(ScalarValue::Int32(None)),
                     distinct_count: Precision::Exact(0),
                     null_count: Precision::Exact(0),
-                    byte_size: Precision::Absent,
+                    byte_size: Precision::Exact(0),
                 },
             ]
         );