Fixes for 3dvar_atmos cycling_varbc

[mranst]

Description

This PR fixes a few issues related to setting cycling_varbc to true in 3dvar_atmos. flow.cylc is fixed for this option, and GetObsNotInR2d2 has been adapted to link previous cycle bias files the same way GetObservations does.

I think there was a bug in the way that previous cycle bias files were linked that has been present for a while. The issue was that when determining the files to link, the process would take the current cycle's bias input file and only replace the cycle directory where it was stored. This would result in the file it was looking for having the wrong background time offset. For example, when running the 20251230T060000Z cycle, the file would start as:

root_directory/20251230T060000Z/geos_atmosphere/aircraft_temperature.20251229T210000Z.acftbias_cov 

It would then only replace the cycle directory, so you would get

root_directory/20251230T000000Z/geos_atmosphere/aircraft_temperature.20251229T210000Z.acftbias_cov 

When in fact the filename should be:

root_directory/20251230T000000Z/geos_atmosphere/aircraft_temperature.20251229T150000Z.acftbias_cov 

As this is the previous cycle time minus background time offset for that period, though I could be wrong about how this is supposed to work

abi_g16 is switched off, as it doesn't look like x0054 has observation files for it. Should this normally be a failure state?

[mranst]

#854