Consider using the --memory-mapping flag on kraken2 to avoid high RAM usage

[dfornika]

We've been experiencing some memory-related issues (filled swap space) on one of our systems recently. It's likely (though we haven't proven) that our kraken2 analyses are filling up main memory and spilling over into the swap space.

The way this pipeline is currently configured is quite inefficient in terms of RAM usage. Each job loads the kraken2 database into RAM. When using the standard database, that involves loading ~60 GB of data into RAM for each job.

kraken2 offers a --memory-mapping flag that avoids loading the database into RAM. It will almost certainly be slower, but may still be acceptable.

Allow users to use the --memory-mapping flag on the kraken2 process. Consider which mode should be the default.