internal/xds: xDS v3 server implementation

[stevesloka]

Follows #3016 by adding an xDS v3 server. Follow-up PR will introduce a config option to specify the xDS version wanting to serve.

Updates #1898

[codecov]

Codecov Report

Merging #3017 into main will decrease coverage by 0.29%.
The diff coverage is 53.33%.

@@            Coverage Diff             @@
##             main    #3017      +/-   ##
==========================================
- Coverage   74.30%   74.00%   -0.30%     
==========================================
  Files          92       95       +3     
  Lines        6008     6081      +73     
==========================================
+ Hits         4464     4500      +36     
- Misses       1446     1479      +33     
- Partials       98      102       +4     

Impacted Files	Coverage Δ
internal/xds/v3/server.go	0.00% <0.00%> (ø)
internal/xds/v3/contour.go	69.09% <69.09%> (ø)
internal/xds/resource.go	100.00% <100.00%> (ø)
internal/xds/v2/contour.go	89.09% <100.00%> (-0.20%)	⬇️
internal/dag/cache.go	95.36% <0.00%> (-0.78%)	⬇️

[jpeach]

Do we really need to duplicate all the code here? The only think that is really different is the types of the discovery request and response. Can we take an approach where we only have to duplicate or abstract the guts of (*contourServer)stream()?

[stevesloka]

I pulled out some more bits that were duplicated. Please take a look @jpeach, @skriss.

[skriss]

It seems like in order to reuse much more code here, we'd need to define some interfaces and wrapper structs to adapt the v2 & v3 discovery request and response structs to common interfaces that the stream method could then interact with.

Given that we're looking to replace this all with the go-control-plane xds server though, I'm not sure it's really worth spending that effort.

[skriss]

Couple nits but overall I'm fine with this as-is, given that (1) we're going to drop v2 support pretty soon since it will be removed from Envoy early next year; (2) we're in the process of moving exclusively to the go-control-plane impl; and (3) upstream Envoy will not be revving a major version of the API every year going fwd.