We have a class in GalSim config called LoggerWrapper, which is intended to avoid communicating with the main logger via the proxy when we aren't actually going to output anything. E.g. logger.debug statements when the logging level is only INFO or WARNING. This is mostly a big improvement.

However, @beckermr and @jchiang87 have each recently found huge inefficiencies when running on Perlmutter with ~128 processes. I tracked it down to when we make a new LoggerWrapper instance and call logger.getEffectiveLevel(). This had always been going back to the original proxy logger instance even when we already have a wrapped one. And for lots of processes, this can end up clogging the inter-process communication completely unnecessarily.

This PR makes the fairly trivial change that when we're re-wrapping something that is already wrapped, it doesn't call getEffectiveLevel. It just gets the already received value stored as an attribute. This seems to resolve the blockage that Matt (and I presume Jim) was seeing.