kvserver: let BenchmarkEmptyRebalance wait for the raft group to be ready

For leader leases, the raft group will take a few seconds before it can
propose new entries because it needs store liveness heartbeats to start.
This commit modifies the benchmark by making it wait until the group is
ready.

Benchmark results:
name                          old time/op    new time/op    delta
EmptyRebalance/add-remove-12     2.85s ± 1%     0.03s ± 8%  -98.96%  (p=0.000 n=7+8)

name                          old alloc/op   new alloc/op   delta
EmptyRebalance/add-remove-12    12.3MB ±13%     1.9MB ± 3%  -84.85%  (p=0.000 n=10+8)

name                          old allocs/op  new allocs/op  delta
EmptyRebalance/add-remove-12     64.7k ± 2%     10.8k ± 0%  -83.29%  (p=0.000 n=10+8)

Fixes: #144940

Release note: None

Author: iskettaneh

pkg/kv/kvserver/client_replica_test.go

@@ -5795,6 +5795,17 @@ func BenchmarkEmptyRebalance(b *testing.B) {
 	defer tc.Stopper().Stop(ctx)
 
 	scratchRange := tc.ScratchRange(b)
+
+	// Before actually starting the benchmark, we need to make sure that the raft
+	// group is able to add/remove voters. This is important because in leader
+	// leases, it takes a few seconds for store liveness heartbeats to start.
+	// We need store liveness heartbeats for two reasons: (1) By default,
+	// followers won't campaign unless they are supported by a quorum of peers,
+	// and (2) The leader won't be able to propose config changes unless the new
+	// config doesn't cause a regression in the LeadSupportUntil.
+	tc.AddVotersOrFatal(b, scratchRange, tc.Target(1))
+	tc.RemoveVotersOrFatal(b, scratchRange, tc.Target(1))
+
 	b.Run("add-remove", func(b *testing.B) {
 		b.ResetTimer()
 		for i := 0; i < b.N; i++ {