fix(orchestrator): re-trigger publishes need distinct message IDs

[behinddwalls]

The mysql message queue makes publishes idempotent on the (topic, partition_key, id) unique key by documented design (safe publish retries), and rows are only removed by offset-based garbage collection — so a publish reusing a prior message ID is silently coalesced against any not-yet-collected row, including already-consumed ones. The new speculation pipeline reuses IDs in exactly the places that must RE-trigger work, and the Phase-B respeculation e2e run caught all of them at once: prioritize rounds published under the bare queue name meant only the first round per queue ever ran (a batch becoming Speculating a moment later was never promoted, never built, and waited forever), and the dependency-failure fan-out published under the dependent's bare batch ID — the same key the score stage had already used — so the wake-up vanished and the dependent stayed stuck at scored.

Every re-trigger publish now mints an ID that is unique exactly when a new round/wake-up is needed and coalescing when it is not. Prioritize rounds: {queue}/{batch}/v{treeVersion} — every persisted tree change is guaranteed a round (and every prioritizer input change is some tree's version bump), while no-change passes coalesce. Dependent fan-out: {dependent}/dep-terminal/{dep} — one wake-up per (dependent, terminal dependency), self-heal redeliveries coalesce among themselves. Prioritize's build republish: {batch}/v{treeVersion} — promotion rounds are guaranteed delivery, pure heal republishes coalesce.

The terminal self-heal branch now re-fans dependents for every terminal state, not just Cancelled/Failed: Succeeded is how dependents learn a dep landed and dead-path the siblings that bet against it — mergesignal already publishes the terminal batch to speculate on success, so this branch is precisely its landing spot.

✅ make fmt && make test (79/79) — speculate suite updated for the new IDs (per-test versions derived from each test's own Update expectations) and for Succeeded's dependent re-fan. The Phase-B respeculation e2e (next in stack) is the regression test that caught both drops: it goes from permanently stuck at scored to landing.

Stack

1. feat(speculate): write the speculation tree in shadow mode #353
2. feat(speculation): route builds through the tree via prioritize #365
3. feat(speculation): reconcile dead paths and finalize from the tree #371
4. feat(speculation): cancel batches through the tree via prioritize #379
5. feat(speculation): optimistic merge finalization behind a confirmed hand-off #380
6. test(e2e): assert speculation-tree parity on the landing pipeline #381
7. @ fix(orchestrator): re-trigger publishes need distinct message IDs #382