Add save_freq to CAPQL.train()

[AleNimo]

When training the CAPQL agent, I noticed a drop in the simulation speed when setting checkpoints = True. In the code, if checkpoints = True, the weights of the nets are stored after every step taken by the agent. A parameter such as save_freq could be added to save the net's weights only every save_freq number of global steps.

Another improvement would be to save the weights in different files as its being done in the GPI-PD agent, for example.

[LucasAlegre]

Thanks for spotting this! I will open a PR to fix it :)