[flang] Fix folding of EPSILON()

[klausler]

Summary

The value of EPSILON() was off by one power of two due to my
misreading of the standard; this patch corrects it.
Confirmed by checking other Fortran compilers, which all agree.

[PeteSteinfeld]

Looks great!

flang/include/flang/Evaluate/real.h

template <typename INT> constexpr INT EXPONENT() const {
     } else {
       return {UnbiasedExponent()};
     }
   }
 
   static constexpr Real EPSILON() {
     Real epsilon;
     epsilon.Normalize(
-        false, exponentBias - binaryPrecision, Fraction::MASKL(1));
+        false, exponentBias + 1 - binaryPrecision, Fraction::MASKL(1));
     return epsilon;
   }
   static constexpr Real HUGE() {
     Real huge;
     huge.Normalize(false, maxExponent - 1, Fraction::MASKR(binaryPrecision));
     return huge;
   }
   static constexpr Real TINY() {

flang/test/Evaluate/folding07.f90

integer, parameter :: &
     bs16 = bit_size(0_16)
   logical, parameter :: test_bit_size_1 = bs1 == 8
   logical, parameter :: test_bit_size_2 = bs2 == 16
   logical, parameter :: test_bit_size_4 = bs4 == 32
   logical, parameter :: test_bit_size_8 = bs8 == 64
   logical, parameter :: test_bit_size_16 = bs16 == 128
 
   real(2), parameter :: &
-    eps2 = epsilon(0._2), zeps2 = real(z'1000', kind=2), deps2 = 4.8828125e-4_2
+    eps2 = epsilon(0._2), zeps2 = real(z'1400', kind=2), deps2 = 9.765625e-4_2
   real(3), parameter :: &
-    eps3 = epsilon(0._3), zeps3 = real(z'3b80', kind=3), deps3 = 3.90625e-3_3
+    eps3 = epsilon(0._3), zeps3 = real(z'3c00', kind=3), deps3 = 7.8135e-3_3
   real(4), parameter :: &
-    eps4 = epsilon(0._4), zeps4 = real(z'33800000', kind=4), &
-    deps4 = 5.9604644775390625e-8_4
+    eps4 = epsilon(0._4), zeps4 = real(z'34000000', kind=4), &
+    deps4 = 1.1920928955078125e-07_4
   real(8), parameter :: &
-    eps8 = epsilon(0._8), zeps8 = real(z'3ca0000000000000', kind=8), &
-    deps8 = 1.1102230246251565404236316680908203125e-16_8
+    eps8 = epsilon(0._8), zeps8 = real(z'3cb0000000000000', kind=8), &
+    deps8 = 2.2204460492503130808472633361816406250e-16_8
   real(10), parameter :: &
-    eps10 = epsilon(0._10), zeps10 = real(z'3fbf8000000000000000', kind=10), &
-    deps10 = 5.42101086242752217003726400434970855712890625e-20_10
+    eps10 = epsilon(0._10), zeps10 = real(z'3fc08000000000000000', kind=10), &
+    deps10 = 1.08420217248550443400745280086994171142578125e-19_10
   real(16), parameter :: &
     eps16 = epsilon(0._16), &
-    zeps16 = real(z'3f8e0000000000000000000000000000', kind=16), &
-    deps16 = 9.629649721936179265279889712924636592690508241076940976199693977832794189453125e-35_16
+    zeps16 = real(z'3f8f0000000000000000000000000000', kind=16), &
+    deps16 = 1.9259299443872358530559779425849273185381016482153881952399387955665588378906250e-34_16
   logical, parameter :: test_eps2 = eps2 == zeps2 .and. eps2 == deps2
   logical, parameter :: test_eps3 = eps3 == zeps3 .and. eps3 == deps3
   logical, parameter :: test_eps4 = eps4 == zeps4 .and. eps4 == deps4
   logical, parameter :: test_eps8 = eps8 == zeps8 .and. eps8 == deps8
   logical, parameter :: test_eps10 = eps10 == zeps10 .and. eps10 == deps10
   logical, parameter :: test_eps16 = eps16 == zeps16 .and. eps16 == deps16
 
   integer(1), parameter :: &