Inconsistent instrumentation semantics

Hi,

I'd like to point out, and ask, what's the expected behavior when an exception is thrown in a Grape endpoint, towards instrumentation (via AS:N).

When I want to instrument Grape, I subscribe to this notification:

https://github.com/ruby-grape/grape/blob/master/lib/grape/endpoint.rb#L238

What I'm experiencing right now is that endpoint.status is always 200 (as opposed to 500 or default_error), even when an exception was raised in that endpoint.

When an exception is raised, I guess this blows up:

https://github.com/ruby-grape/grape/blob/master/lib/grape/endpoint.rb#L264

But then what happens with the instrumentation block?

When explicitly stating an error inside an endpoint, it works (status code is 500).
In a rescue_all block, or while an endpoint throws it doesn't work (status code is either stale, wasn't set, and is just 200)

Start in lib/grape/endpoint.rb at the instrumentation notification near line 238 and the exception path near line 264. Reproduce explicit errors, rescue_all handling, and uncaught endpoint exceptions, then compare the reported status values; done means the instrumentation semantics are confirmed and status reporting is consistent across these paths.