feat: move-stable row ids in compaction (#2544)

Part of #2307.

Also addresses #2397

rust/lance-table/src/rowids.rs

@@ -125,7 +125,7 @@ impl RowIdSequence {
  self.0.extend(other.0);
  }
 
- /// Mark a set of row ids as deleted. Their value will be replaced with tombstones.
+ /// Remove a set of row ids from the sequence.
  pub fn delete(&mut self, row_ids: impl IntoIterator<Item = u64>) {
  // Order the row ids by position in which they appear in the sequence.
  let (row_ids, offsets) = self.find_ids(row_ids);
@@ -152,6 +152,27 @@ impl RowIdSequence {
  self.0.extend_from_slice(remaining_segments);
  }
 
+ /// Delete row ids by position.
+ pub fn mask(&mut self, positions: impl IntoIterator<Item = usize>) -> Result<()> {
+ let row_ids = positions
+ .into_iter()
+ .map(|pos| {
+ self.get(pos).ok_or_else(|| {
+ Error::invalid_input(
+ format!(
+ "position out of bounds: {} on sequence of length {}",
+ pos,
+ self.len()
+ ),
+ location!(),
+ )
+ })
+ })
+ .collect::<Result<Vec<_>>>()?;
+ self.delete(row_ids);
+ Ok(())
+ }
+
  /// Find the row ids in the sequence.
  ///
  /// Returns the row ids sorted by their appearance in the sequence.

rust/lance-table/src/rowids/segment.rs

@@ -390,8 +390,6 @@ impl U64Segment {
  })
  };
  let stats = Self::compute_stats(make_new_iter());
-
- // Then just use Self::From_stats_and_sequence
  Self::from_stats_and_sequence(stats, make_new_iter())
  }
 }