feat: latency-based failover

[mzz2017]

Background

I want to treat my US node as main use, and if it is down, failover to HK node. In this way, US nodes with high latency can be used preferentially.

Idea

Normalize all failover situations to latency-based solution.

This solution has following advantages:

1. If the US node is under DDOS, the connectivity is not stable and the node will be up and down. In this situation, a timeout (10000ms) will be added to the latency every down and this node will have lower priority. The priority will recover automatically if DDOS is over.
2. Reuse current policies: min, min_avg10, min_moving_avg, ...

group {
  my_group {
    filter: name(my_HKnode) [add_latency: 300ms]
    filter: name(my_USnode, node_c) [add_latency: -500ms]
    filter: name(node_d, node_e, node_f)
    policy: min_avg10
  }
}

Thus, we need to introduce the annotation grammar([add_latency: 50ms]).

Checklist

• The Pull Request has been fully teste
• There's an entry in the CHANGELOG
• There is a user-facing docs PR against https://github.com/daeuniverse/dae

Full changelog

• Support annotation syntax for declaration ([key: value, ...]).
• Support multiple filters in one group.
• Support add_latency annotation for filter and apply it to corresponding dialers (if multiple filter hits, only the first works).

Issue reference

Fix #118

[mzz2017]

Some changes need to do. The added latency should be in group level.

[mzz2017]

One more thing:
It changed behavior that "no filter" means "no filter" to "no filter" means "no node". It should be fixed.

[mzz2017]

"Group level 'add_latency'" and "no filter problem" have been fixed. Ready for review.

[mzz2017]

Test result:

[dae-prow]

🧪 Since the PR has been fully tested, please consider merging it.

[mzz2017]

Another test result: