Relocate batch merging traits into the fueled spine

[frankmcsherry]

What

Moves the Merger trait and the Batch::Merger associated type out of trace/mod.rs and into spine_fueled.rs, as a spine-local trait:

• SpineBatch: Batch (in spine_fueled.rs) associates a batch type with its Merger. The Rc forwarding (RcMerger, impl SpineBatch for Rc<B>) moves alongside it.
• Batch (in trace/mod.rs) retains only empty, which has a consumer outside the spine: TraceWriter::seal mints empty batches to pad otherwise empty intervals of time, so empty-minting is a property of writable traces generally, not a spine opinion.
• Implementors (OrdValBatch, OrdKeyBatch, ChunkBatch) split into a one-method impl Batch and a separate impl SpineBatch.

Batch::begin_merge does not survive the move. It and Merger::new were two entry points to the same construction, and both implementors overrode begin_merge with the trait's own default body. Only Merger::new remains; the documentation of what beginning a merge means moves onto Merger. The spine's own MergeState::begin_merge, which handles the structurally-empty cases, is unaffected.

No behavior changes; all other edits are path and name updates (including three test modules).

Why

An audit showed the merge machinery had exactly one consumer: the fueled spine. Different trace maintenance strategies want differently shaped merge contracts (compare merge_batcher's own Merger, which owns its input chains), so the progressive-merge signature is an opinion of this spine rather than a property of batches in general. Relocating it makes later changes to that opinion — e.g. merge state that owns its inputs and releases them incrementally, to avoid the 2x memory spike during merges of chunk batches — local to the spine instead of revisions to the common batch contract.

Verification

cargo build --workspace (and --examples), cargo test --workspace --lib, and cargo test -p differential-dataflow --tests all pass; cargo doc introduces no new warnings.

🤖 Generated with Claude Code