ENH: Produce "small" metadata.tsv with only essential columns

[corneliusroemer]

Context

Our metadata.tsv is now so big that it becomes impossible to load into memory in full.

We should either create an "essential" metadata.tsv or in general partition metadata.tsv and make users merge later.

Right now we sort of require preprocessing with tsv-select to pick relevant columns for interactive work (though augur filter seems to work fine with the metadata file as is so this isn't urgent).