Imbalance with crowd size in optimizer cost function

[markdewing]

In #3873, this issue was raised and deferred to a future change.

In QMCCostFunctionBatched checkConfigurations and correlatedSampling, the number of samples is divided by the number of crowds, then by the number of walkers. It would lead to less imbalance if the number of samples were divided by the number of walkers first, then the number of crowds (to make the number of samples per walker more even).

Prior to #3873, the number of walkers per crowd was the same for every crowd. As part of the change to match VMC, the number of walkers per crowd may not be the same for every crowd.

If the number of walkers is evenly divisible the number of crowds, this is not an issue.