Log samples per second in base_agent.py

[mwulfman]

I’m proposing a small addition to _log_train_info to log samples per second.

I’ve found this metric useful as a quick way to monitor training throughput. It makes it easier to spot when a code change may need optimization, and to estimate how long a run will take to collect the same number of samples as another experiment.

New log train info output:

------------------------------------------
|            Iteration |             737 |
|            Wall_Time |           0.981 |
|              Samples |        48365568 |
|   Samples_Per_Second |        1.16e+04 |
|          Test_Return |             165 |
|         Train_Return |             174 |
|                 Loss |           0.214 |
|          Critic_Loss |           0.141 |
|           Actor_Loss |          0.0722 |
|            Clip_Frac |           0.697 |
|            Imp_Ratio |            0.99 |
|    Action_Bound_Loss |        4.77e-07 |
|             Adv_Mean |            0.38 |
|              Adv_Std |           0.341 |
|             Exp_Prob |               1 |
------------------------------------------