DADA2 memory optimizations and bugfixes

[m-murphy]

Functionally everything is the same, this commit reduces peak memory usage and improves readability. Outputs should not change.

• No longer preload dereplicated files, reducing peak memory usage. DADA2 automatically dereplicates files as needed
• Fix function calls to respect args$cores input
• remove large objects from memory as they are finished being used, reducing peak memory usage especially at end of script
• rewrite the unique allele merging function to be more readable

[bgpalmer]

this looks good to me