Term increment is skipped in one-node clusters

[cole-miller]

When the current configuration contains only one node, we have a fast path in raft to skip the election and move directly from follower to leader:

dqlite/src/raft/convert.c

Lines 171 to 180 in 6035a35

	/* Fast-forward to leader if we're the only voting server in the
	* configuration. */
	server = configurationGet(&r->configuration, r->id);
	assert(server != NULL);
	assert(server->role == RAFT_VOTER);
	if (n_voters == 1) {
	tracef("self elect and convert to leader");
	return convertToLeader(r);
	}

This is great, but because the election code is responsible for incrementing the term number, we end up skipping this update for one-node clusters as well. This means we can have two distinct leaders that share a term number, as happens in this test:

dqlite/test/integration/test_cluster.c

Line 127 in 6035a35

TEST(cluster, dataOnNewNode, setUp, tearDown, 0, cluster_params)

I think this behavior is unintuitive at best and we should add a term number increase on the fast path.