An idle build machine's empty fetch is asked again, not read as the end (hq ADR 0190)

A fetch on a context without a deadline waits the client's own while and reports the deadline
passed — the client's, not ours — and the loop read it as "stop": every idle build agent exited
clean every half minute and was restarted by its supervisor, a crash loop with nothing in the log
to say why. Only our own context ending ends the machine; an empty fetch, however it is reported,
is asked again.

internal/link/builds_nats.go

@@ -193,9 +193,15 @@ func (m *natsMachine) Take(ctx context.Context, do func(context.Context, Build))
 		// machine with nothing to build, and is asked again.
 		fetched, err := sub.Fetch(1, nats.Context(ctx))
 		switch {
-		case errors.Is(err, context.Canceled), errors.Is(err, context.DeadlineExceeded):
+		case ctx.Err() != nil:
+			// Ours ended: the machine is being stopped.
 			return nil
-		case errors.Is(err, nats.ErrTimeout):
+		case errors.Is(err, context.Canceled), errors.Is(err, context.DeadlineExceeded), errors.Is(err, nats.ErrTimeout):
+			// **An empty queue, not the end.** A fetch on a context without a deadline waits the
+			// client's own while and then says the deadline passed — the client's, not ours. Read
+			// as "stop", every idle build machine exited clean every half minute and was started
+			// again by its supervisor, which looked like a crash loop with nothing in the log to
+			// say why (2026-10-03, the first build agents). Asked again.
 			continue
 		case err != nil:
 			if sub.IsValid() {