Split ParallelBest into "client groups" and "parallel"

[ThinkChaos]

The ParallelBestResolver fulfills two purposes:

1. choosing upstreams based on the client
2. querying those in parallel

I think we should split it.

The way I'd try to implement it is by having each group's list of resolvers actually be a single "parallel" resolver.
From an architecture point of view, blocky would be less "list of resolvers" and more tree-like. This change was already somewhat started with #449.

Going towards a tree-like architecture would also be a good step towards #476.

I have no plans of actually working on this at the moment, just want to track this and get feedback.

[0xERR0R]

I think it is a good point to think about different resolving strategies. Currently, we pick 2 random resolvers (random-weight based on last error timestamp) and take result from the fastest. We could also support following (just some examples):

• use one resolver (round robin)
• use always first resolver. If resolution not successful, take second and so on
• rank resolvers by response time, take fastest
• ....

[0xERR0R]

relevant issue: #355

[github-actions]

This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue was closed because it has been stalled for 5 days with no activity.

[0xERR0R]

@ThinkChaos: is there some open points after #1093 was merged?