Removed clocks that were being called from inside of j-loops in two modules.
These are inefficient and can cause the model to hang in some cases if used, and
there are better ways to get timing information at this level.  If there is
interest in the timing breakdown at this level, the code should be restructured
to move the key blocks outside of the j-loops.  The run-time parameter
ALLOW_CLOCKS_IN_OMP_LOOPS is no longer being used so it is now obsoleted. All
answers are bitwise identical, but there are changes to some MOM_parameter_doc
files.

  Added a deallocate call for eta_PF_start in step_MOM_dyn_split_RK2() to
avoid a possible memory leak.  All answers are bitwise identical.

  Set find_salt_root even if SHELF_THREE_EQN = .False. to avoid using an
uninitialized logical to determine which parameters are logged.  Without this
the contents of some MOM_parameter_doc.all files could depend on the state of
uninitialized memory and was compiler dependent in some cases.  All answers are
bitwise identical, but in some cases the contents of MOM_parameter_doc files
could be corrected.

  The runtime parameter ETA_TOLERANCE_AUX was being read but was never used, so
it is being obsoleted.  However, because some experiments were using this and
there are effectively no changes in behavior, a warning will be issued instead
of a fatal error if this parameter is set.  All answers are bitwise identical,
but there are changes to some MOM_parameter_doc files.