Add updateAttributes as Span protocol requirement

[adam-fowler]

Add updateAttributes(_:) as Span protocol requirement
The majority of Span implementations will use the default implementation
Adding the function to the protocol though allows us to override the function for the NoOpSpan

Fix for #189

[ktoso]

LGTM, this works fine because the default impl already exists; so this will be source compatible, which is the only guarantee we need here;

Therefore we can disregard the API break warning, it doesn't understand the default impl

[czechboy0]

Hmm see my other comment, I still think we should call the closure, just with local empty span attributes that get discarded immediately, not saved on the instance.

[adam-fowler]

We then allocate space for attributes ie the ones we setup inside the closure and then throw them away immediately. It appears on godbolt the compiler isn't clever enough to compile out the creation of the attributes.

[czechboy0]

My lean would be that we just live with that cost and file an issue on the compiler to improve this.

But when correctness and performance are in tension, and we're locking this decision in a public API, I think we should choose correctness here.