augur curate I/O: verify NDJSON records have the same fields

[joverlee521]

Context

Originally discussed in https://github.com/nextstrain/augur/pull/1506#discussion_r1660441721

augur curate records can be output to a metadata TSV file, which uses the first record's fields as output columns.

https://github.com/nextstrain/augur/blob/f6ee377336ec3813468fe641fa7910c14e54ced3/augur/io/metadata.py#L467-L474

With that in mind, the centralized inputs parser should verify that the input records all of the same fields. Then subcommands can operate under the assumption that all records should have the same fields and make changes accordingly.

[jameshadfield]

the centralized inputs parser should verify that the input records all of the same fields

Probably sensible to do this for outputs too, to assert the curate subcommand's run() always returns a consistent set of fields. I haven't checked if csv.DictWriter orders each row based on the information previously passed to writeheader() or if it relies on the insertion order being the same for each row, but if it's the latter then we should assert that as well.

[jameshadfield]

Some more details about TSV writing via our generalised write_records_to_tsv:

• We don't need to worry about the order of TSV rows -- DictWriter will re-order each of them to match the header information provided in the set-up.
• Fields missing from a row will be ignored (empty string used in their place). This is also noted in our docstring but it seems from the discussion in this issue that we may want to explicitly assert that each row has the same keys within augur curate.
• Extra fields in a row are currently ignored - also noted in our docstring - but we may want to revisit this and assert it upstream within augur curate.