add timestep bonus to loss

Project/Assets/ML-Agents/Examples/Tennis/Scripts/HitWall.cs

     void AgentAWins()
     {
         m_AgentA.SetReward(1 + m_AgentA.timePenalty);
-        m_AgentB.SetReward(-1);
+        m_AgentB.SetReward(-1 - m_AgentB.timePenalty);
         m_AgentA.score += 1;
         Reset();
 
     {
-        m_AgentA.SetReward(-1);
+        m_AgentA.SetReward(-1 - m_AgentA.timePenalty);
         m_AgentB.SetReward(1 + m_AgentB.timePenalty);
         m_AgentB.score += 1;
         Reset();

config/trainer_config.yaml

     time_horizon: 1000
     self_play:
         window: 10
-        play_against_latest_model_ratio: 0.1
+        play_against_latest_model_ratio: 0.5
         save_steps: 50000
         swap_steps: 50000
         team_change: 100000