alabaster 0.7.16 failed the CI; the last passing test used 0.7.13 so I am
pinning that version

There are also updates to the driver so that the call_compute_subroutines test
still runs when abio_dic_on = T and base_bio_on = F. I suspect there will be
baseline failures in that case due to changes in the output variables

output_for_GCM_flux_co2, output_for_GCM_total_Chl, and
output_for_GCM_total_surfChl are no longer generated by the
call_compute_subroutines test when using marbl_with_abio_only.settings so they
have been removed from the baseline

The registry makes it easier to add new outputs for the GCM or to add more
restrictions on when certain outputs are available while maintaining useful
error messages.

Also cleaned up some white space / alignment issues

This required a lot of small changes.

MARBL library:

1. Rename surface_flux_output* -> output_for_gcm*
2. Add total_Chl index to indexing type, and rename sfo_ind -> ofg_ind
3, Remove marbl_instance%get_output_for_GCM()
4. Add field_source to registry type as well as marbl_single_output_type (so
   GCM knows whether the field should be copied after surface_flux_compute() or
   after interior_tendency_compute())
5. Generalize registry definition, and make more specific error messages
   (flux_o2 requires both base_bio_on and lflux_gas_o2 to be true)
6. Pass output_for_GCM to interior_tendency_compute so it can populate
   total_Chl (if requested by GCM)

Stand-alone driver:

1. rename total_Chl -> interior_tendency_output, add num_vars dimension (this
   is the 3D equivalent to surface_flux_output). Note that there is an
   assumption that all outputs_for_GCM from surface_flux_compute are 2D, while
   all from interior_tendency_compute are 3D. This is fine for now, but we may
   need to introduce datatypes if future mods contradict that assumption
2. Rely on field_source in output_for_GCM type to know what to copy out of the
   type / when to copy it
3. Total Chl is no longer a special case in marbl_io_mod.F90, it comes through
   the output_for_GCM object

Put surface_flux_output back on the interface, and also created
interior_tendency_output. This change lets us drop field_source from the output
type, and add_output_for_GCM() does all the error checking I had introduced
into marbl_single_output_constructor (making sure the variable is registered
and available).

Note that add_output_for_GCM() now returns field_source as well, so the GCM can
keep the surface flux output and interior tendency output separate when copying
to local memory (without knowing what fields come from surface_flux_compute()
vs interior_tendency_compute())

In surface_flux_mod.F90, I had changed the argument to
marbl_surface_flux_compute() from output_for_gcm to surface_flux_output (this
is what it was called before I tried to integrate the outputs for
surface_flux_compute and interior_tendency_compute; I've split those outputs
apart again, so the name change is not necessary)

1. Better comments in create_registry
2. The registry error message is now prepended with the field name in
   add_output_for_GCM()
3. create_registry can use base_bio_on, doesn't need it as an intent(in)
4. better comment in add_output_for_GCM (logs an error message, doesn't return
   one)
5. If ofg_ind is a target, components of output_for_gcm_type don't need to be
   pointers (registry%id => ofg_ind%* works)

Still have a few slightly bigger changes to make in the next commit

Replace marbl_output_for_GCM_single_register_type with
marbl_output_for_GCM_linked_list_type; array of former becomes a pointer to
latter. This will make it much easier to add a new output_for_GCM, simply copy
format of a block of code in create_registry() [and add entry to
marbl_output_for_GCM_indexing_type]

it is called from create_registry, which is already in the
marbl_output_for_GCM_registry_type class

1. Typo in marbl_interface.F90 -- abort if registered_output is not associated
   (was checking this%output_for_gcm_registry%registered_outputs by accident)
2. Add comment about why create_registry() resets err_message to "" before each
   entry (and clean up errant white-space misalignment)
3. Also, blocks in create_registry() are easier to read if we set err_message
   to "" and then use a continutation line for the if () rather than using
   if-else
4. For adding to the registry linked list, order doesn't matter but the logic
   is simpler (and code is faster) if we prepend entries rather than looping
   through to the end of the list every time

Provides a mechanism to see what output_for_GCM fields are available for a
given configuration. Note that output_for_gcm_registry is now a public member
of the MARBL interface class so the standalone test can access it.

A new page explains what additional output is available from the GCM, and a
second page shows how to add a new output. The new available_output test is
also documented on the regression tests page.