Reuse cursor(sample) for all agents per trial

[sroecker]

Since there is no need to use a different precalculated sample for each agent the cursor can be updated in the environment after each trial.

[bgalbraith]

Hi @sroecker. I think what you're addressing is the fact that when we do choose to pre-cache random samples in the bandit, we should draw from the same iteration sample pool for every agent instead of each agent getting a unique set of samples per iteration. That's a valid point, though introducing coupling between the environment and the bandit, especially to an ostensibly private property, isn't the best solution.

Thanks for the input!