feat(stovepipe): migrate process and buildsignal waits to the hold primitive

[behinddwalls]

Summary

Why?

The process stage's build-slot wait and buildsignal's status poll loop were built on ack-and-republish: each wait tick acked the delivery and PublishAfter'd a fresh copy of the same message to the controller's own topic. That forced both controllers to be publishers to their own topics, required per-stage message-id minting (a wall-clock suffix in process, a deterministic generation counter in buildsignal) to dodge the queue's (topic, partition_key, id) dedup, wrote a new log row per tick, and hung each loop's liveness on a publish succeeding. The hold primitive (doc/rfc/consumer-hold.md) is the framework-owned replacement.

What?

process: the budget-full branch now records a hold for GateWaitDelayMs and returns success; the delivery is threaded through processAccepted/admitLatestHead, rescheduleProcess and its wall-clock message-id minting are deleted, and the positive-delay config guard moves into holdForBuildSlot. buildsignal: a non-terminal status holds the delivery for the per-status poll delay; nextPollMessageID, the /poll/ generation scheme, and publishBuildSignal are deleted, and publish loses its now-unused delay parameter. The held message is a partition barrier that redelivers in order without counting toward the retry limit, so the DLQ budget stays reserved for real failures, and a failed postpone write lapses into a normal visibility-timeout redelivery instead of killing the loop.

Docs updated to the decided mechanism: process.md's "Waiting for a slot" section replaces the previously-deferred Option 1/Option 2 comparison with hold (the primitive that deferral named), buildsignal.md's polling-primitive and algorithm sections describe hold, GateWaitDelayMs is re-documented as a redelivery delay, and the e2e slow-build test's obsolete dedup rationale is rewritten.

Test Plan

✅ bazel test //stovepipe/... — reworked unit tests assert Hold(delay) is recorded on the gate-wait and poll paths and that no Hold happens on terminal/error paths. ✅ bazel test //test/e2e/stovepipe/... — the slow-build scenario reaches a terminal build through multiple hold-driven poll ticks and releases the build slot. ✅ make fmt.

Stack

1. docs(rfc): consumer hold — postpone a delivery as a partition barrier #486
2. feat(platform): hold/postpone — a fourth delivery outcome for backing off #487
3. @ feat(stovepipe): migrate process and buildsignal waits to the hold primitive #488
4. feat(orchestrator): migrate the buildsignal poll loop to the hold primitive #489
5. feat(messagequeue)!: remove PublishAfter and the visible_after column #490
6. feat(messagequeue)!: drop the requeue delay parameter from Nack #491
7. feat(consumergate)!: rebuild the gate on the postpone primitive #492