name_resolution: include structural attributes in Address PartialEq

[nathanielford]

This PR is for the grpc-Rust official library, and not for tonic.

Motivation

In the process of writing the Pick First LB policy, I realized that there were edge cases characterized by the following:

• The name resolver provides a new list of addresses
• Some of those addresses match to the previous list of addresses based on the hashed fields, even though they should not because they have different attribute fields, causing a false positive match.

The reason for this is that the addresses cannot be included in the hashing, because they are type-erased. I think all load balancers are likely to run into this issue.

Solution

Updating PartialEq in order to validate that the address attributes are equal in addition to the network type and address field. This means new Addresses with different attributes should be hashed into the same bucket but that looking an address with attribute set A will not fetch an address with attribute set B.

There may be a better way to handle this, or a reason it hasn't been done already - I'd be grateful for any feedback that anyone can provide on this front.

[dfawley]

I don't recall if there is a reason it was like this before, but this SGTM.

Can we not derive(PartialEq,Eq)? (If not why not?)

[nathanielford]

No reason not to use derive! I just didn't think of it right off since I was in the weeds. Have fixed that. Also included a perhaps not needed unit test in order to at least capture the thinking. If we feel it's not worth the bytes, though, I can pull it.