Add conflict-compatible commutative transaction merges

[xav-db]

Summary

• Add the opt-in DbTransaction::merge_commutative API, with the inherent Rust method as the canonical contract and matching trait and UniFFI guidance.
• Classify transaction writes as exclusive or commutative merges and omit same-key write/write conflicts only when both transactions contain exclusively commutative merges for that key.
• Preserve serializable point/range read dependencies, mixed-operation conflict behavior, ordinary merge semantics, and unmark_write.
• Cover compatible merges, ordinary merge/put/delete conflicts, mixed writes, repeated classification, SSI reads, and existing untracked-write behavior.

Why

SlateDB currently treats blind merge-only updates like exclusive writes. That forces retries on hot keys even when the configured merge operator is order-independent, such as counters, set union, or bitmap union.

The new API makes that stronger algebraic contract explicit. Both operands remain in the normal atomic write batch; only conflict metadata changes, and callers remain responsible for commutativity.

Impact

Applications can opt compatible same-key merge operands out of unnecessary write/write conflicts without weakening read tracking or changing persistence. Existing callers and ordinary merges retain their current conservative behavior.

Validation

• All-feature workspace tests and doctests
• Clippy with warnings denied
• Rustfmt check
• Targeted branch tests for every write-classification path

cargo llvm-cov executed the targeted tests successfully, but its final report exporter terminated with SIGSEGV in the local environment.