Increase tolerance for one continuation test

The conserved energy quantity for the largest system for
the mdrun continuation test, alanine_vacuo, already had
a higher tolerance for single precision. Now a uniform
increase is used. This needs a higher tolerance because
the conserved energy is relatively small due to the sum
of (negative) potential energy and (positive) kinetic energy.

Fixes #4725

src/programs/mdrun/tests/exactcontinuation.cpp

@@ -433,9 +433,13 @@ TEST_P(MdrunNoAppendContinuationIsExact, WithinTolerances)
         if (simulationName == "alanine_vacuo")
         {
             // This is slightly less reproducible
-            energyTermsToCompare.insert({ interaction_function[F_ECONSERVED].longname,
-                                          relativeToleranceAsPrecisionDependentUlp(
-                                                  10.0, ulpToleranceInMixed * 2, ulpToleranceInDouble) });
+            // Wider tolerance is needed because of sum of negative PE
+            // with positive KE producing small conserved energy.
+            const int factor = 4;
+            energyTermsToCompare.insert(
+                    { interaction_function[F_ECONSERVED].longname,
+                      relativeToleranceAsPrecisionDependentUlp(
+                              10.0, factor * ulpToleranceInMixed, factor * ulpToleranceInDouble) });
         }
         else
         {