linear_batch doc needed in the manual

[ye-luo]

I failed to find opt_crowd_size in the manual.

[ye-luo]

Also about how opt_num_crowds is determined in the current implementation.

[ye-luo]

This is my real-world user experience with the batched WFOpt driver. @markdewing could you help with this?

[markdewing]

• opt_num_crowds is an input parameter. If not set (or set to zero), it is set to the number of threads, Concurrency::maxCapacity<>();, same as the other batched drivers.
• opt_crowd_size is an input parameter - set to 1 by default. This is the number of samples processed together when evaluating parameter derivatives.

In my testing, the setting of opt_crowd_size didn't have much effect on the performance. That was a year ago, though, so maybe the offload evaluation has improved since them.

[ye-luo]

We need to document the current behavior and then think about better deafult.

opt_num_crowds and opt_crowd_size if not give, better to ask VMC what is its decision.

In my testing, the setting of opt_crowd_size didn't have much effect on the performance. That was a year ago, though, so maybe the offload evaluation has improved since them.

~~Need high walker count to see the effect but I do see 4x difference in time in checkConfiguraitons.~~ See https://github.com/QMCPACK/qmcpack/issues/3849#issuecomment-1045451893 10x difference