Consolidate the emitted-carrier producer-binding consumers onto one seam

ProducerBinding has three emitted-carrier arms (DmlEmitted, ServiceEmitted, and the routine arm that roadmap/routine-mutation-payload-carrier-return.md adds) sharing a consumer-facing shape: reflectedClass, tableRef, arrival, the correlation columns, and an identical reflectedClass.getName().equals(tableRef.recordClass().reflectionName()) compact-constructor invariant. Their genuine difference is provenance, consumed only by describe() and the multi-producer rejection.

The carrier-return item lands the EmittedCarrierBinding capability over the three arms, so the shared accessors are read once and totally. What it deliberately leaves alone is the consumer side, which by then carries the same structure three times over: three memo maps on RecordBindingResolver, three xEmittedBinding accessors, and three near-duplicate blocks in FieldBuilder.classifyChildFieldOnResultType (resolve binding, resolve return type, lift a polymorphic errors field, check the table agreement, call buildPayloadCarrierBatchedTableField). No consumer forks on the arm’s identity; every one reads the same two or three accessors.

Two sibling sites are not in this item’s scope, because the carrier-return item has to touch them to work at all: TypeBuilder.carrierBinding’s per-family probe (without a routine probe the payload never registers as a carrier) and `FieldBuilder.transportForParent’s `activeChannel disjunction (without a routine disjunct the carrier’s errors field binds Transport.PayloadAccessor, which a directiveless structural carrier cannot satisfy). They arrive already three-armed; if the capability gained the presence probe that gate needs, the disjunction is already gone by the time this item starts.

Scope: fold the memo maps and accessors onto the capability, collapse the classify-time blocks to one, and unify the table-agreement diagnostics without losing the per-family wording that existing fixtures pin (the DML block’s "payload-returning DML mutations require child @table-bound fields to bind to the input table" and its siblings). Only two diagnostics are real: the routine block’s agreement is tautological at its seat (its binding’s table is read off the same data field the check would compare it to), so that block is a thin duplicate by design. The diagnostics are the real work; the structure is mechanical.

Deliberately not folded into the carrier-return item: that item touches neither the DML nor the @service emit path, and pulling this in would put both families' emit into its acceptance surface for a cleanup neither needs.