Golay error decoding for the EMP protocol

[wasade]

This pull request adds in Golay error decoding for the Earth Microbiome Project protocol (solving #72). The code included is essentially a direct copy from QIIME1, which was originally authored by @justin212k, with a few exceptions outlined below:

• some normalization of data types and improved utilization of numpy
• methods wrapped in an object
• the use of functools.lru_cache to avoid decoding unless necessary
• standardized the docstrings

The contained code does more than only decoding, so it is possible to use the included library methods for the purpose of creating new Golay codes. However, since demux only presently supports the EMP protocol, it didn't seem necessary at this time to expose any of that additional functionality.

As this code was brought from an existing project, some additional detail such as the module docstring was retained.

Finally, this PR does introduce an API change to both emp-single and emp-paired. Specifically, we now record detail about the error correction process. This information is dumped into a DataFrame and provided as an output. The original intention was for this output to be compatible with qiime metadata tabluate, however the sample-id column is now unique (as multiple reads within a sample may have errors) so it does not currently work with that visualizer.

[wasade]

One more note, without the lru_cache, decoding roughly doubled runtime when demuxing the MVP tutorial dataset. With the lru_cache, there is no appreciable added runtime (anecdotal).

[thermokarst]

Quick pass today, food for thought. I will do a more in-depth review soon.

[thermokarst]

Lookin' good @wasade, made one more pass. Next time through I want to take a closer look at tests. Otherwise looking good! Thanks! 💯

[thermokarst]

👍

cc @ebolyen - this format would be another good candidate for a schema-enabled format. @wasade, defining these tabular formats is cumbersome, and is on our laundry-list of things to do around here. No need to change anything here.

[thermokarst]

How about you add the header/type mapping as a class property to this format? Then, you can use it here and here (and in the rest of the transformers).

[thermokarst]

Third comment on this format - can you make this a TSV? If you did, then you could actually just use the qiime2.Metadata object for validation:

for column in self.METADATA_COLUMNS.keys():
    try:
        md.get_column(column)
    except ValueError as md_exc:
        raise ValidationError(md_exc) from md_exc

As well, your transformers could use qiime2.Metadata for loading and saving.

[wasade]

Good call!

I got in the header / type mapping stuff. I was less sure about how to do the changes to support the Metadata object though, is there a recommended example to base off of by chance?

[thermokarst]

There is a little bit here in this open PR: qiime2/q2-types#210

[thermokarst]

Okay, I have just a few minor things to add to this, then it is set:

• Test when a read is skipped due to >3 errors
• Test the _validate_ behavior on ill-defined files
• What happens if barcodes are not 12 nts long and ec is enabled?

Thanks!