Hold every kept archive by a manifest so the store's collector keeps it (hq issue 253)

The store's garbage-collect marks only from manifests, and archives were
published as bare blobs, so the first real collection would delete every
archive the mesh keeps. PublishArchive now puts a deterministic OCI holder
manifest (empty config, one layer) beside each archive; the sweep holds every
kept archive before it lets anything go, which backfills existing bare blobs,
and lets go of an archive holder-first. A forgotten module no longer keeps its
five recent builds (ADR 0189). `collection [--json]` reports kept archives
held/unheld and what may be let go, so the dry run can be lifted on evidence.

cmd/mesh-controller/collect.go

@@ -31,7 +31,12 @@ func collect(ctx context.Context, inv *inventory.Inventory) {
 		fmt.Fprintf(os.Stderr, "could not work out what the artifact store may let go of: %v\n", err)
 		return
 	}
-	if len(references) == 0 {
+	kept, err := inv.KeptArchives(ctx)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "could not work out which archives the artifact store keeps: %v\n", err)
+		return
+	}
+	if len(references) == 0 && len(kept) == 0 {
 		return
 	}
 	shelf, err := inv.Catalogue(ctx)
@@ -59,6 +64,25 @@ func collect(ctx context.Context, inv *inventory.Inventory) {
 	defer stop()
 	store := artifacts.Store{Address: address}
 
+	// **Hold before letting go** (novox/hq issue 253). The store's collector keeps only what a
+	// manifest names, and archives were published as bare blobs, so every kept archive is first
+	// held by its manifest — which backfills the ones published before holders, a few at a time
+	// as builds come, and is two HEADs each once done. A kept archive that could not be held stops
+	// the sweep before it deletes anything: "everything kept is held" is the precondition the
+	// collector's safety rests on, and a store refusing a hold would refuse the deletes too.
+	wrote, missing, err := holdKept(within, store, kept)
+	if wrote > 0 {
+		fmt.Fprintf(os.Stderr, "the artifact store now holds %d more kept archive(s) by a manifest\n", wrote)
+	}
+	if missing > 0 {
+		fmt.Fprintf(os.Stderr, "%d archive(s) the mesh keeps are not in the artifact store at all; "+
+			"`collection` lists them\n", missing)
+	}
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "not every kept archive could be held, so nothing was let go: %v\n", err)
+		return
+	}
+
 	var done []string
 	var left, skipped int
 	for i, reference := range references {
@@ -114,6 +138,34 @@ func collect(ctx context.Context, inv *inventory.Inventory) {
 	}
 }
 
+// holdKept holds every kept archive by its manifest, stopping at the first refusal by the store.
+// Answers how many holders it wrote and how many kept archives the store does not have.
+//
+// A missing archive is counted rather than fatal: there is nothing to hold, and that is a fact
+// for an operator to read (`collection`), not a reason to stop collecting what is not kept. A
+// reference the store cannot be asked about is skipped as the deletion loop skips one
+// (novox/hq issue 226).
+func holdKept(ctx context.Context, store artifacts.Store, kept []string) (wrote, missing int, err error) {
+	for _, reference := range kept {
+		if err := ctx.Err(); err != nil {
+			return wrote, missing, fmt.Errorf("ran out of time before %s: %w", reference, err)
+		}
+		did, err := store.Hold(ctx, reference)
+		switch {
+		case err == nil:
+			if did {
+				wrote++
+			}
+		case errors.Is(err, artifacts.Gone):
+			missing++
+		case errors.Is(err, artifacts.ErrNotOurs):
+		default:
+			return wrote, missing, fmt.Errorf("holding %s: %w", reference, err)
+		}
+	}
+	return wrote, missing, nil
+}
+
 // mostPerSweep is how many artifacts one sweep will ask about. Enough that a mesh building
 // several times a day converges within days of this landing; small enough that no single build
 // waits on the whole backlog.