Feat request: write out obs_seq files

[hkershaw-brown]

The capability to write obs sequences from the dataframe opens up preprocessing of observations for dart in the python ecosystem, e.g.

• subsetting by:

  • observation type, qc, ... attributes of the observation
  • time
  • area (clicking on a map)

• preprocessing:

  • removing observations above/below a height

Potentially replacing functionality from these various fortran programs:

• changing-observation-sequence-files
• querying-observation-sequence-files
• creating-observation-sequence-files

Issues:

• individual obs_defs are not self describing in the obs_seq file. It is metadata without sufficient info (refer to as incomplete-metadata?)
  • you can store the incoming obs_def incomplete-metadata for a given obs, but for creating observations from scratch you would need to call the fortran observation routines with the current design of DART. (or extract the complete metadata from the fortran for each obs_def, e.g. obs_def.yaml)

@hkershaw-brown this would make creating obs_seq.{final/out} files easier for tests.

[hkershaw-brown]

For dataframes where observations have been removed (or even had their time altered), the linked list info needs to be updated.

I don't think we have to manage the linked list, we can:

• sort dataframe by time
• number observations by row+1
• linked list info is then always the same -1 2 -1 1 3 -1 2 4 -1 ...

[hkershaw-brown]

still need metadata storage and writing