Request memories for post processing hard coded

Currently request memories for post processing are hard coded to be relatively large values (up to 32GB) in the codes. The large request memories often cause troubles for online PE running on MDC events, since post processing jobs do not find nodes to run on and pile up. Also, looking at the log of online PE, they actually use a very small fraction of requested memories (less than ~1GB).

Thus, it will be useful to have additional arguments in parser to enable users to specify those memories, so that we can reduce those values for test usages such as MDC runs. @gregory.ashton @colm.talbot Any thoughts on this? I am happy to open a small MR to do this, if this idea sounds fine.