Change AthenaMP ChunkSize to match input auto-flush, unless shared reader is used.

This change will have default AthenaMP dispatch chunks of events equal to auto-flush when reading LZMA compressed data (most data PxAOD and up) to avoid duplicating de-compression work. For Run3 this is most applicable for derivation, which will be run in AthenaMP on LZMA data with 100 events per cluster. I closed the previous MR as this one is needed to address problems with workers in very short jobs (e.g. testing) not getting any events dispatch and stopping the shared writer. Also this doesn't use chunk size for shared reader, as there is no double decompression in that case.