Make co_heat test numerically stable & stricter

Due to an overly fine grid and an explicit time step, the co_heat
test was previously unstable.  Furthermore, due to a lax test,
previously reported results were false positives.   This commit
increases the grid spacing, thereby improving stability and adds
additional checks on the result. My best guess on why this issue
is just cropping up is that GCC 8 must be much slower on arithmetic
involving IEEE NaN values so the test now fails for all attempted
values of the ctest --timeout argument.

Author: Damian Rouson

src/tests/integration/pde_solvers/coarrayHeatSimplified/main.f90

@@ -26,12 +26,13 @@
 !
 
 program main
+  use IEEE_arithmetic, only : IEEE_is_NaN
   use global_field_module, only : global_field
   implicit none
   type(global_field) :: T,laplacian_T,T_half
   real, parameter :: alpha=1.,dt=0.0001,final_time=1.,tolerance=1.E-3
   real :: time=0.
-  call T%global_field_(internal_values=0.,boundary_values=[1.,0.],domain=[0.,1.],num_global_points=16384)
+  call T%global_field_(internal_values=0.,boundary_values=[1.,0.],domain=[0.,1.],num_global_points=32)
   call T_half%global_field_()
   do while(time<final_time)
     T_half = T + (.laplacian.T)*(alpha*dt/2.)
@@ -40,6 +41,10 @@ program main
   end do
   call laplacian_T%global_field_()
   laplacian_T = .laplacian.T
-  if (any(laplacian_T%state()>tolerance)) error stop "Test failed."
+  block 
+    real, allocatable :: residual(:)
+    residual = laplacian_T%state()
+    if ( any(residual>tolerance) .or. any(IEEE_is_NaN(residual)) .or. any(residual<0) ) error stop "Test failed."
+  end block
   if (this_image()==1) print *,"Test passed."
 end program