Fixed G%gridLoaTn and G%girdLonB in "spherical" model

- The diagnostic axes "xh" and "xq" were mostly filled with zeros
  when the grid was "spherical" (not tri-polar, etc.) due to two
  [quite understanable] typos in the loop ranges that set up G%gridLonT
  and G%gridLonB.
- Corrects axes info in netcdf files.
- Closes NOAA-GFDL/MOM6-examples#39.
- No timestats changes.

src/initialization/MOM_grid_initialize.F90

@@ -1041,10 +1041,10 @@ subroutine set_grid_metrics_spherical(G, param_file)
     latitude = G%south_lat + dLat*(REAL(j-G%jsg)+0.5)
     G%gridLatT(j) = MIN(MAX(latitude,-90.),90.)
   enddo
-  do i=G%IegB,G%IegB
+  do i=G%IsgB,G%IegB
     G%gridLonB(I) = G%west_lon + dLon*(REAL(I-(G%isg-1)))
   enddo
-  do i=G%ieg,G%ieg
+  do i=G%isg,G%ieg
     G%gridLonT(i) = G%west_lon + dLon*(REAL(i-G%isg)+0.5)
   enddo