Avoid associate statements using index of 0?

[mnlevy1981]

In both set_surface_forcing and its associated diagnostics routine, we have large associate statements that include blocks of code pointing at specific elements of arrays. For example

    associate(                                                                                  &
         ind_diag          => marbl_surface_forcing_diag_ind,                                   &
         ind_forc          => surface_forcing_ind,                                              &

         diags             => surface_forcing_diags%diags(:),                                   &
         xkw               => surface_input_forcings(:,surface_forcing_ind%xkw_id),             &
         xco2              => surface_input_forcings(:,surface_forcing_ind%xco2_id),            &
         xco2_alt_co2      => surface_input_forcings(:,surface_forcing_ind%xco2_alt_co2_id),    &
         ap_used           => surface_input_forcings(:,surface_forcing_ind%atm_pressure_id),    &
         ifrac             => surface_input_forcings(:,surface_forcing_ind%ifrac_id),           &
         dust_flux_in      => surface_input_forcings(:,surface_forcing_ind%dust_flux_id),       &

As we move to a more flexible codebase, it is possible that some of these surface_forcing_ind elements will be 0 (if, for example, we want to run without the "alternate CO2" computations, surface_forcing_ind%xco2_alt_co2_id=0). I thought we ran into an issue at one point where a compiler complained about associating to something out of the bounds of the array, but I don't remember the specifics... so I'm creating this issue ticket as a place to leave comments if it comes up again.

I believe @klindsay28 is working on a branch of the code where surface_forcing_ind%iron_flux_id=0 - can you please leave a comment and let us know if any of the big three compilers (Intel, GNU, NAG) complains?

Two possible work-arounds:

1. Nested associate statements (would need to see how compilers react to that, as well)
2. Pointers instead of associate statements (would need to set up target attribute in a lot of data types)