Add restart capability to the fire behavior model

[masih-e]

Description of changes:

The CFBM has had no way to stop a simulation and resume it.
This feature adds a restart capability. A run can write its dynamic fire state to a
NetCDF file at a fixed interval, and a later run can be initialized from one of
those files instead of a cold start. It covers standalone runs, serial and MPI,
and runs coupled to WRF.

Two options are added to the &time namelist block:

Option	Type	Default	Meaning
restart	logical	.false.	Resume from fire_restart_<start datetime>.nc
restart_interval	integer	-1	Restart write interval in seconds; -1 disables writes

Both default to the previous behavior, so existing namelists are unaffected.

Changes by file:

• state/state_mod.F90 carries the Read_restart and Write_restart
  type-bound methods, Build_restart_file_name, and the MPI helpers they need.
  A restart read validates the grid metadata stored in the file against the
  current configuration before touching any state, so a mismatched restart file
  fails early with a message instead of producing a silently wrong run.
• io/netcdf_mod.F90 adds attribute writers (Add_netcdf_att) and
  MPI-aware readers and writers for distributed 2D and 3D fields.
• driver/initialize_mod.F90 routes the real-case restart path through a
  new Init_fire_state_from_restart, leaving the existing geogrid
  initialization untouched. The WRF-coupled path reads the restart after the
  fire components are initialized.
• driver/fire_behavior.F90 reads the restart state at startup in place of
  Save_state, and writes a restart every restart_interval seconds.
  restart_interval must map to a whole number of time steps or the run stops
  with a message.
• io/namelist_mod.F90 adds the two options, their broadcast, and a range
  check.
• physics/fmc_wrffire_mod.F90 drops its unused state_mod and
  namelist_mod imports. This one is load-bearing rather than cosmetic:
  state_mod now uses fmc_wrffire_mod for the moisture restart fields, so
  leaving those imports in place would create a circular module dependency.

Type of change

• New feature (non-breaking change which adds functionality)

Tests conducted:

All testing on Derecho (NCAR) using the repository environment
env/derecho/gnu-12.2.0: GCC 12.2.0, cray-mpich 8.1.25, NetCDF-Fortran 4.9.2,
HDF5 1.12.2, ESMF 8.5.0, CMake 3.26.3.

Builds. Three configurations, all clean with no errors or new warnings:
serial (--mpi-off), MPI (default), and --esmx (which includes NUOPC).

Registered ctest suite, run on the --esmx build so every registered case
is reachable:

test7	test8	test7esmf	test8esmf	test7esmx	test8esmx	testx
pass	pass	pass	pass	pass	pass	pass

7 of 7. The NUOPC cases matter here because the cap uses both initialize_mod
and state_mod, and this PR touches both.

Restart regressions. These run a case straight through, then run it again
stopping and restarting partway, and compare the two. Restart test scripts are
not included in this PR (see below), but the results are:

Serial build, 20 of 20 checks:

test7_restart	test8_restart	test9	test9_restart
5/5	5/5	5/5	5/5

MPI build on a compute node, 30 of 30 checks:

ranks	test7_restart	test8_restart
1	5/5	5/5
2	5/5	5/5
4	5/5	5/5

The multi-rank runs are the ones that exercise the distributed restart read and
write and the halo exchange following a restart read. Restarting at 2 and 4
ranks reproduces the continuous run, so the domain decomposition is handled
correctly on the restart path.

57 passing checks in total across the three build configurations.

Not covered. Two gaps, stated explicitly:

1. The WRF-coupled restart path (Init_fire_state_within_wrf calling
   Read_restart) is exercised by a separate WRF harness, and not tested here.
2. Rank counts above 4 have not been tried.

On the test scripts. The restart test scripts (test7_restart.s,
test8_restart.s, test9.s, test9_restart.s, the test9 case directory, and
the tests/CMakeLists.txt registration) are deliberately held out of this PR
pending the separate discussion about how tests should be organized. They are
ready and can be added here or in a follow-up PR, whichever is preferred.

Documentation:

• Documentation has been updated for these changes
• No documentation update needed

Documentation was left out to keep the diff focused on
the code. Happy to add it to this PR if preferred, otherwise it will follow
separately.

Contributors:

@masih-e (Masih Eghdami:masih@ucar.edu)
@mefrediani (Maria Frediani:frediani@ucar.edu)

LLM/coding assistant technology used: the restart implementation was developed
with assistance from Codex GPT-5.5 and Claude Opus 4.7. Claude Opus 5 assisted
with the pre-review cleanup of this branch.

Further information:

state/state_mod.F90 grows from 1063 to 1965 lines in this PR, which is more
than that file should carry. A restart-specific module is worth doing as a follow-up. This
change is not possible with the current code because the restart calls depend on the fire
state type.