fix RepeatedFieldTest.Enumerator test

[jtattermusch]

on my Windows machine RepeatedFieldTest.Enumerator is failing with error:

Test FullName:  Google.Protobuf.Collections.RepeatedFieldTest.Enumerator
Test Source:    c:\Users\jtattermusch\github\protobuf\csharp\src\ProtocolBuffers.Test\Collections\RepeatedFieldTest.cs : line 240
Result Message: System.InvalidOperationException : Operation is not valid due to the current state of the object.
Result StackTrace:  
at Google.Protobuf.Collections.RepeatedField`1.Enumerator.get_Current() in c:\Users\jtattermusch\github\protobuf\csharp\src\ProtocolBuffers\Collections\RepeatedField.cs:line 506
at Google.Protobuf.Collections.RepeatedFieldTest.Enumerator() in c:\Users\jtattermusch\github\protobuf\csharp\src\ProtocolBuffers.Test\Collections\RepeatedFieldTest.cs:line 246

it looks like all changes to the enumerator variable in that test are immediately forgotten.
I eventually found the cause, which seems to be:
http://stackoverflow.com/questions/4642665/why-does-capturing-a-mutable-struct-variable-inside-a-closure-within-a-using-sta
and this bug report
https://connect.microsoft.com/VisualStudio/feedback/details/635745

Just removing the using block fixes the problem (as well are removing the lambda would), but it makes me wondering if we actually want RepeatedField.Enumerator to be a struct. Exposing enumerator struct makes as susceptible to subtle issues like these:
https://www.simple-talk.com/blogs/2010/05/19/why-enumerator-structs-are-a-really-bad-idea/
https://www.simple-talk.com/blogs/2010/05/20/why-enumerator-structs-are-a-really-bad-idea-redux/

(I think with Reset() method we are actually fine, because we don't implement IEnumerator.Reset() explicitly, unlike System.Collections.Generic.List, but I am wondering how much of a performance gain the struct-based enumerator actually is.

Switching the enumerator to private class might save us some headache in the future (MapField already has an enumerator class).

[jskeet]

The struct-based enumerator can provide a significant win, particularly if the list is small or empty.

Now I'm not sure we're using foreach over the list anywhere any more - we can validate that.

[jtattermusch]

To fix the test, we basically have these options
-- remove the using statement
-- stop using the lambda in the test
-- change the iterator type to class

Which one do you prefer?

[jskeet]

Will see if changing to use try/finally explicitly fixes it instead, which I think I'd prefer. But will investigate the current benefits in the runtime, too.

[jskeet]

Replaced with #610