Total chlorophyll is set to 0 initially, and is not updated until after the
first time MOM6 calls get_chl_from_model. This causes two problems:
1. We do not want to compute swpen based on 0 chlorophyll in the first time
step
2. Restart tests to fail, because total chlorophyll was being reset back to 0
in the first time step of the restart

There are two related fixes in this commit:
1. All interior tendency / surface flux outputs from MARBL are now in the
restart file, so the model passes ERS tests
2. If total chlorophyll was not initialized from a restart file, it is computed
based on the initial values of the MARBL tracers (which could be from an IC
file or a restart file)

The dummy cap did not contain a component of marbl_single_output_type that I am
now using in MARBL_tracers.F90

Instead of passing G -> register_MARBL_tracers(), can use HI to determine i,j
extent of CS%SFO and CS%ITO. Also, change order of k and m loops for better
memory management (and condense double-loops to a single line)