Simplified parallel API

[RagnarGrootKoerkamp]

Instead of a dedicated mapper struct, in cases where on_batch_complete is not needed, a much simpler version that directly takes a closure may be sufficient?

Something like, reader.process_parallel(|record| { ... }).

But I suppose one benefit of the current approach is to accumulate per thread, in which case the mapper struct seems inevitable.

[noamteyssier]

Yeah that was my thinking - I originally wanted to implement it as a closure like seq_io but ended up deciding against it as I found it too inflexible.

What I like about the binseq ParallelProcessor trait is that I can effectively mirror it with the paraseq ParallelProcessor trait so that you can write the per-record logic once and get both binseq and fastx for free