Allow exceptions without HystrixRuntimeException wrapper

[tbvh]

Introduce NotWrappedByHystrix to mark Exceptions that should not be wrapped in HystrixRunetimeException.

Partial fix for #1283 .

[mattrjacobs]

Thanks for the PR @tbvh ! This looks like a good change to me. I'm trying to think through side-effects of this change.

A) Only users who opt into using this interface get a change in behavior - that's a good thing
B) At runtime, the effect of the change is that executions that return any Exception instance which implements ExceptionNotWrappedByHystrix gets the raw exception
C) These exceptions manifest as FAILURE in the FailureType enum and metrics

For C), would it make sense to have a new event type (UNWRAPPED_FAILURE) that would make it easier to distinguish this scenario?

Any other users in the community have thoughts on this change?

[Sunshow]

So will this PR be merged to master?

[tbvh]

I just posted #1283 (comment).

Just to clarify, I've introduced this as a partial fix, but it completely solves point 1 of issue #1283. In the comment I propose to drop point 2, which make this PR in fact a complete fix.

Makes me all the more curious to the merge status. :)

[mattrjacobs]

A couple thumbs-up from the community and I agree with the change. I think it's fine not to special-case this for metrics. A failure of the wrapped work should be a FAILURE as far as metrics are concerned, whether the exception is passed back directly or now. If anyone disagrees with this, then I think that' fine to have in a follow-on PR.