MRG: enable singleton sketching, facilitate reads-based sketching

[bluegenes]

Here I modify manysketch to allow:

• sketching multiple input files into the same sketch
• sketching singletons!

New Functions

• detect_csv_type - figure out which type of acceptable input csv we have
• process_assembly_csv - process standard fromfile csv (headers: name,genome_filename,protein_filename). This was originally part of load_fasta_fromfile.
• process_reads_csv - process new reads csv (headers: name,read1,read2)

This refactoring also makes it much simpler for us to add new types of CSVs that can be read.

Notes:

• parallelization is still done by each set of inputs that should yield signatures. If singleton, we're sketching all records for a given file in series, not all sketches in parallel.
• perhaps a confusing thing: when sketching with --singleton, the name entry in each row doesn't end up getting used, since we name from the records themselves.

1. Multiple input files for a single sketch

We can now sketch multiple filenames into the same sketch. This is currently only possible if you input a 'reads' csv, with format name,read1,read2, which is designed to add support for metagenomes with PE reads. Note that the read2 column name must be present, but can be empty. Since this is designed for metagenome reads, we assume input files are DNA.

The original sketch fromfile input format, name,genome_filename,protein_filename remains functional.

Future work could introduce additional ways to pass more than one filename per sketch.

2. Singleton sketching

With either input filetype, we can now pass --singleton and get one sketch per record, named from the record name.

Prefix-based sketching punted to #243

[bluegenes]

@ctb ready for review.

[ctb]

Looks great, works great!

Some minimal documentation and an update to the help docstring would be lovely :). But punt to an issue if you prefer to add those later.