feat(tuple): serialize union results without cloning summaries #321

Problem

Tuple union serialization currently requires union.to_sketch(ordered).serialize(). to_sketch() clones retained entries, including each user-defined summary. For summaries containing heap data, this may allocate and copy substantial data that serialization only needs to borrow. The summary encoder itself already accepts &self.

Suggested approach

• Add direct union-result serialization that borrows summaries while applying the final theta, capacity, and ordering rules.
• For ordered output or capacity selection, consider sorting/selecting references to entries instead of cloning summary values. Share result preparation and encoding with the compact sketch where appropriate.
• Keep the existing TupleSummaryValue encoding contract. Avoid adding a Clone bound solely for serialization; this does not imply removing the cloning requirements of union updates.
• Begin with TupleUnion. Evaluate reuse for TupleIntersection and mutable-sketch compaction/serialization as follow-ups. Preserve the distinction between an uninitialized intersection and an empty result.

Start with datasketches/src/thetafamily/tuple/union.rs, tuple/sketch.rs, tuple/serialization.rs, and common/union.rs. Coordinate changes to shared Theta-family result preparation with the Theta serialization work. PR #317 provides the corresponding HLL motivation.

Validation

Compare with the existing compact-result path across ordering choices, theta/capacity trimming, empty/singleton results, and fixed- and variable-size summaries. Exercise summary encoding and subsequent merges with a realistic heap-backed summary. Verify existing compatibility tests and measure time and allocations for both small numeric summaries and larger heap-backed summaries before claiming a performance gain.

Follow CONTRIBUTING.md for changelog entries, integration-test placement, serialization fixtures, and the required checks.