Switched to my atomic-approach for the placements.db

[jgolob]

This is just a different way to accomplish the same task. I liked it because it leveraged SCONS to at least let me know if / when one of the steps in making the placements.db failed.

This particularly matters with guppy (at least the publicly available version) being flaky (in which a placements.db is created on the filesystem, but guppy dies during the classification step).

[nhoffman]

I'd be game for separating out the guppy classify step from other steps. However, the boundaries that you draw between other steps seem artificial since multicalss_concat.py is already doing a whole bunch of stuff within the script - why not just extend that script to include ingesting the adcl and seq_info tables? I'm actually hoping to rely less on the database. I am going to dig into what multiclass concat is actually doing - reduplication in the database seems redundant since the dada2 inputs already provide weights. I think that the initial design anticipated that the database would be used directly for all sorts of analyses, but the reality has been that most manipulations happen in pandas or R downstream. So it may not even be worthwhile to perform any of the steps after guppy classify, and instead extract a single table with classifications of individual sequence variants from the database, and perform additional manipulations elsewhere.