2.1: batch_size as machine option

Created by: gcarleo

Samplers should just define a (maximum) batch size as an option (something to be set up with machine.batch_size=128 or similar). See also #300 (closed).

In turn this means that every time log_val is called on an arbitrary batch size, machines internally divide the call into sub batches. Currently this behavior is implemented only for calls to local energy estimators, but it should be extended (considerably simplifying the code that does it) to all calls of log_val, der_log etc. This will also make things more efficient when calling batched versions of der_log when computing the derivatives in the new python drivers.