[flang] Improve error messages about overflowed integer conversions (llvm#110031)

When an INTEGER conversion to a smaller kind overflows in constant
folding, report the truncated value so that it makes more sense later if
it shows up in other messages.

Author: klausler

flang/lib/Evaluate/fold-implementation.h

@@ -1736,8 +1736,9 @@ Expr<TO> FoldOperation(
                   msvcWorkaround.context.languageFeatures().ShouldWarn(
                       common::UsageWarning::FoldingException)) {
                 ctx.messages().Say(
-                    "INTEGER(%d) to INTEGER(%d) conversion overflowed"_warn_en_US,
-                    Operand::kind, TO::kind);
+                    "conversion of %s_%d to INTEGER(%d) overflowed; result is %s"_warn_en_US,
+                    value->SignedDecimal(), Operand::kind, TO::kind,
+                    converted.value.SignedDecimal());
               }
               return ScalarConstantToExpr(std::move(converted.value));
             } else if constexpr (FromCat == TypeCategory::Real) {

flang/test/Evaluate/errors01.f90

@@ -192,6 +192,8 @@ subroutine warnings
     real, parameter :: bad10 = product([huge(1.),huge(1.)])
     !CHECK: warning: PRODUCT() of COMPLEX(4) data overflowed
     complex, parameter :: bad11 = product([(huge(1.),0.),(huge(1.),0.)])
+    !CHECK: warning: conversion of 111111111111111111111_16 to INTEGER(8) overflowed; result is 430646668853801415
+    integer(8), parameter :: bad12 = int(111111111111111111111, 8)
     !CHECK: warning: overflow on REAL(8) to REAL(4) conversion
     x = 1.D40
     !CHECK-NOT: warning: invalid argument