Reflection: Fix FormatException when querying ParameterInfo.DefaultValue of optional DateTime parameter

[stakx]

Summary:

Fixes a reflection bug (as discussed in https://github.com/dotnet/corefx/issues/26164) which causes a FormatException when calling ParameterInfo.DefaultValue for an optional DateTime parameter having an encoded default value of null (which is how the Roslyn C# compiler encodes default(DateTime)).

Example of the bug being fixed:

The following C# program currently throws an exception:

class Program
{
    public void Method(System.DateTime arg = default(System.DateTime)) { }

    static void Main()
    {
        var argDefaultValue = typeof(Program).GetMethod("Method").GetParameters()[0].DefaultValue;
        System.Diagnostics.Debug.Assert(object.Equals(null, argDefaultValue));
    }
}

Unhandled Exception: System.FormatException: Encountered an invalid type for a default value.
   at System.Reflection.MdConstant.GetValue(MetadataImport scope, Int32 token, RuntimeTypeHandle fieldTypeHandle, Boolean raw)
   at System.Reflection.RuntimeParameterInfo.GetDefaultValueInternal(Boolean raw)
   at System.Reflection.RuntimeParameterInfo.GetDefaultValue(Boolean raw)
   at Program.Main()

This PR makes the above program succeed.

Some additional notes:

• Reflection reporting a default value null when it's default(DateTime) in source code may seem surprising, but is in line with reflection's behavior for how C# encodes other default(<any value type>).

• _{I'm including a test project tests/src/reflection/OptionalParameters/DefaultValue.csproj. tests/src/reflection/OptionalParameters/DefaultValue.ilproj, however it doesn't currently get built automatically (see Refine test discovery and partition during test build with new target #17331). I've run it manually to verify this bugfix.}

[ghost]

I'm including a test project tests/src/reflection/OptionalParameters/DefaultValue.ilproj

could it be a csproj, with the example you posted above? my understanding is that usually things are tested with ilproj, which otherwise can't be test with c# grammar?

[stakx]

@kasper3: A .csproj would work just as well, but it would make the test dependent on an implementation detail of the Roslyn compilers (i.e. that they encode default(TValueType) default parameter values as a null constant).

I'm happy to change the tests to a .csproj.

[stakx]

@atsushikan - Could you please review this?

(I am also submitting additional tests for this to CoreFX. Shall I remove the test project here completely?)

[ghost]

As with the other one, we shouldn't include the test in coreclr, just corefx. Otherwise, LGTM

[stakx]

@atsushikan - removed all test code & rebased.