Fixed a bug: LocalEnsembleDA background datetime

[metdyn]

Description

I find the local ensemble DA workflow was using ebkg from x0050.atmens_ebkgx.20231009_15z.tar, and the KF solver output is geos.analysis.mean.20231009_210000z.nc4. I talked with @rtodling and get a fix here. The correction includes:

• background: datetime: set to the middle of time window
• let window_type be 3D
• x0050.atmens_ebkgx.20231009_21z.tar replaces x0050.atmens_ebkgx.20231009_15z.tar
• eliminated an unused extra background: date: '2023-10-09T21:00:00Z', which appears to non-existent in fv3-jedi ctest input as well.

The generated jedi_localensembleda_config.yaml contains 6-hr 3D-Var-like time stamps.

time window:
  begin: '2023-10-09T21:00:00Z'
  end: '2023-10-10T03:00:00Z'
  bound to include: begin

background:
  members from template:
    template:
      datetime: '2023-10-10T00:00:00Z'
      filetype: cube sphere history
      provider: geos

After running LETKF, the analysis.mean and mean-inc output file names show 0Z stamp.

Provide a detailed description of what this PR does.

Dependencies

Impact