Ask every failed build of the tier on a retry, not only the first (issue 457)

An outcome arriving after its plan failed is kept only in the build records,
so a retry read from the plan alone asked one failed build and the records
then failed the plan again on the next. Settle the tier from the records first.

cmd/mesh-controller/plan_retry.go

@@ -288,6 +288,15 @@ func retryPlan(ctx context.Context, open *stores, id string) (string, error) {
 				"those walks answer them; a newer merge, or `rebuild <module>`, builds again", p.ID)
 		}
 	}
+	// Settled from the build records before anything is judged (novox/hq issue 457): a build of the tier
+	// that failed after the plan did is as failed as the one that failed it.
+	if p.Tier < len(p.Tiers) {
+		recorded, byID, err := recordsOfAsked(ctx, inv, &p, p.Tiers[p.Tier])
+		if err != nil {
+			return "", err
+		}
+		toRetry(&p, recorded, byID)
+	}
 	if err := retryRefusal(p, plans); err != nil {
 		return "", err
 	}
@@ -525,3 +534,16 @@ func retryTierWhole(ctx context.Context, open *stores, p *inventory.Plan, again
 func sendAgain(s *inventory.PlanModule) {
 	s.First, s.FirstAt, s.Gate, s.GatedBy, s.Previous, s.Why = nil, nil, nil, "", "", ""
 }
+
+// toRetry is the modules a retry asks again: every one of the tier that failed, the plan's state
+// settled from the build records first (novox/hq issue 457). A plan fails on the first failure in its
+// tier, and an outcome arriving after that finds no open plan to answer — it is kept only in the build
+// records. Read from the plan alone, a retry asked only the build that failed first, and the records
+// then failed the plan again on the next: each failed build of a tier took a retry of its own. What
+// still runs is left asked, and its outcome is the plan's once the retry sets it building.
+func toRetry(p *inventory.Plan, recorded map[string][]inventory.Build, byID map[string]inventory.Build) []string {
+	if p.Tier < len(p.Tiers) {
+		settleFromRecords(p, p.Tiers[p.Tier], recorded, byID)
+	}
+	return failedIn(*p)
+}