fix: use precision-aware default tolerance in f_approx_equal

Single precision (wp=real(4)) has ~7 decimal digits of precision, so
a default tolerance of 1e-10 is below machine epsilon (~1.2e-7) and
f_approx_equal would almost never return true. Use 1e-6 for single
precision builds and keep 1e-10 for double precision.

The wp == single_precision comparison uses integer parameter constants
that the compiler resolves at compile time, so there is no runtime cost.

Author: sbryngelson

src/common/m_helper_basic.fpp

@@ -8,6 +8,7 @@
 module m_helper_basic
 
     use m_derived_types
+    use m_precision_select
 
     implicit none
 
@@ -29,7 +30,11 @@ contains
         if (present(tol_input)) then
             tol = tol_input
         else
-            tol = 1.e-10_wp
+            if (wp == single_precision) then
+                tol = 1.e-6_wp
+            else
+                tol = 1.e-10_wp
+            end if
         end if
 
         if (a == b) then
@@ -58,7 +63,11 @@ contains
         if (present(tol_input)) then
             tol = tol_input
         else
-            tol = 1e-10_wp
+            if (wp == single_precision) then
+                tol = 1.e-6_wp
+            else
+                tol = 1.e-10_wp
+            end if
         end if
 
         do i = 1, size(b)