Produces a single large metadata file keyed off accession with both
strain names and segment-level information encoded therein. The
following commit will group this "long" metadata TSV into a "wide"
TSV where each row represents a strain.

Allows us to avoid another python script.

Original implementation by @joverlee521 in
<#18 (comment)>

Transforms a metadata file with one row per accession into a file with
one row per strain. Where a strain contains sequences for multiple segments this
will group those segments together.

Segment-specific field names (i.e. those not in --common-strain-fields) are modified
to ensure their suffix is "_{segment}". For instance "accession → accession_HA".

Rows are matched on strain name and basic sanity checking is performed when grouping.
Segments with multiple matches for a given strain are dropped, and strains where all
segments have either zero or multiple matches are dropped entirely. Manual resolutions
may be provided via a `--resolutions` YAML which is a list of dictionaries, each with
keys "strain", "accession" and "segment" informing the program which accession (of
multiple) to use. The resolutions here are taken both from exploring the data and
the existing phylo exclude list.

Any disagreement within the "--common-strain-fields" will result in the strain being
dropped, however empty values may be replaced and ambiguous dates may be replaced with
specific ones (where appropriate).

Mismatched field values across segments (e.g. segments disagree on the
'date') are now resolved by choosing the most common occurrence with
the intention they are resolved upstream, as implemented here.

This approach was the third implementation. Initially I resolved
disagreements within `group_segments.py` via a provided resolutions
YAML. After discussion with @joverlee521 we decided this could be better
implemented via `augur curate` and the original implementation here did
this _after_ the segment grouping, however this made it impossible to
distinguish disagreements which will be fixed vs those which won't¹

NOTE: Here we use accession as the ID, however using strain name would
be better going forward as it would reduce the duplication needed in the
current format. We can't (currently) do this in oropouche because strain
names are added _after_ the curate chain runs.

¹ <#18 (comment)>

Updates the files which ingest uploads and makes the corresponding
changes to the phylogenetic workflow. As metadata (and sequences) now
use "strain" as the unique ID a number of simplifications can be made
to the workflow.

There is one regression: the "accession" column no longer exists and
is thus not exported. We'll fix this in a subsequent commit.

This adds back in segment-specific metadata to the Auspice JSON. There
are multiple ways this can be done, each with trade-offs. The approach
employed here leaves the "_{segment}" suffix on the field names.
Alternatively we could remap the metadata file for each `export` call so
that (e.g.) "accession_S" becomes "accession".

To use the new metadata format where we group by strain. Steps to regenerate:

1. Populate 'data/' with metadata & sequences from an ingest run.
2. Subsample this as example data via:

```
augur filter --metadata data/metadata.tsv --group-by country --subsample-max-sequences 30 --output-metadata example_data/metadata.tsv
augur filter --metadata example_data/metadata.tsv --sequences data/L/sequences.fasta --output-sequences example_data/sequences_L.fasta
augur filter --metadata example_data/metadata.tsv --sequences data/M/sequences.fasta --output-sequences example_data/sequences_M.fasta
augur filter --metadata example_data/metadata.tsv --sequences data/S/sequences.fasta --output-sequences example_data/sequences_S.fasta
```