Fix rounding error in getLikelyClassesOrMethods

[EgorBo]

It seems that sometimes getLikelyClasses reports that fallback has e.g. 1% likelihood where in fact it's 0. It happens because of the rounding error accumulated in getLikelyClassesOrMethods. It prevents JIT from moving that fallback to a cold section.

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

It seems that sometimes getLikelyClasses reports that fallback has e.g. 1% likelihood where in fact it's 0. It happens because of the rounding error accumulated in getLikelyClassesOrMethods. It prevents JIT from moving that fallback to a cold section.

public static void Main(string[] args)
{
    for (int i = 0; i < 200; i++)
    {
        Test(new MyClass1());
        Test(new MyClass1());
        Test(new MyClass1());
        Test(new MyClass2());
        Test(new MyClass3());
        Thread.Sleep(16);
    }
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int Test(IValue value) => value.GetValue();

- 1) 00007FFF3995A8C0 (MyClass1) [likelihood:50%]
+ 1) 00007FFF3995A8C0 (MyClass1) [likelihood:51%]
  2) 00007FFF3995AA88 (MyClass2) [likelihood:37%]
  3) 00007FFF3995AC50 (MyClass3) [likelihood:12%]

Author:	EgorBo
Assignees:	EgorBo
Labels:	area-CodeGen-coreclr
Milestone:	-

[AndyAyersMS]

LGTM