Make DeactivationOnIdle behaviour on POCO grain consistent

[rinkeb]

The current behaviour of a call to DeactivateOnIdle depends on whether the grain is a normal grain (deriving Grain) or a POCO grain (implementing IGrainBase). When called on the first a mocked implementation (noop) is called, while on the latter a NotImplementedException is thrown.

In my opinion the choice for either is an implementation detail and should not result in different grain behaviour.

This pull request proposes a change to make the behaviour of the latter consistent with the former.

[seniorquico]

I agree that it should be possible to assert for calls to DeactivateOnIdle. This same issue exists for MigrateOnIdle, as well.

I don't think this is an appropriate interface, though. It shouldn't be a requirement for TestKit consumers to have to cast objects to get to the mock object. In the other path, there's a mock object for IGrainRuntime. I also don't think TestKit consumers should have to figure out which of these deactivation mocks to assert on based on their grain implementation pattern.

[rinkeb]

I agree that it should be possible to assert for calls to DeactivateOnIdle.

Well, for me its not really about asserting. I just want to be able to call DeactivateOnIdle in the SUT without experiencing an exception when the grain happens to be implemented as POCO.

I just mimicked the solution of #9 to have a similar way to validate the behaviour. Not sure about better options, as i'm not that familiar with the code base.

[rinkeb]

Never mind, i migrated to InProcessTestCluster which seems a better fit for my testing needs.