Read renaming process loads all read headers into memory simultaneously

[bede]

While examining the performance of detaxizer on large datasets of 50-100GB in size, I noticed the nextflow process RENAME_FASTQ_HEADERS_PRE using more than 100GB of RAM and repeatedly terminating the pipeline. I see that a dictionary of renamed read headers grows unbounded during renaming, leading to failure with large input files.

In order to use Detaxizer on a machine with 128GB of RAM, I have implemented buffered writing to bound the growth of the renaming dict, and offer my changes as a PR: #62

Process RENAME_FASTQ_HEADERS_PRE is also the most time consuming process in the workflow due to the use of Biopython's relatively slow FASTQ parser. I have replaced this with the >10x faster dnaio parser. I've also accordingly changed this process label to process_low, allowing more renaming processes to run simultaneously for given resource.