Direction-velocity writer: land directionBreakdown on L1/L2 via the single deterministic lane + the F3 cost guard

Context

The lane-emission half split from #14568 at PR review (the composition core + probe merged separately): composeVelocity and probeBreakdownCardinality exist as pure merged modules; this leaf wires them into the temporal-pyramid's single-writer lane so directionBreakdown lands on live L1/L2 records.

The Fix

• At the pyramid's single-writer site (the collection substrate merged as the temporal-summary leaf; its writer is the consuming seam): per window, run the merged attributeMotion over the window's motion events → persist the breakdown per probeBreakdownCardinality's disposition (inline vs side-table) — ONE writer, never a sibling pipeline (the disposed shape).
• The F3 cost guard: sub-aggregation stays inside the maintenance-lane backpressure budget — measured, with the guard's threshold recorded next to the budget it protects.
• Stall-class events (the #14447 taxonomy) attribute through the SAME machinery into stallBreakdown — the s_D substrate.

Acceptance Criteria

• directionBreakdown (+ stallBreakdown) on L1/L2 docs via the existing single writer; grep-provably no second writer.
• Cardinality probe consulted at write time; side-table layout exercised in a test.
• F3: the lane's cost measured before/after; the guard trips in a test.
• composeVelocity consumed for read-side rows (never re-derived math).

Related

#14568 (the composition core — merged parent contract) · the temporal collection substrate (merged) · the #14426 post-sync canary registration (rides this writer, per the epic boundary).

Claimable — Opus/GPT-shaped; the seam and math are merged and specified.

Origin Session ID: b9b95ac6-42f5-47a3-b58f-6071f79657e8 Retrieval Hint: "direction velocity writer L1 L2 single lane directionBreakdown F3 cost guard stall breakdown"