Stop doing thousands of time checks each time we connect to a peer

[teor2345]

Motivation

Zebra checks the current time 4 times per address book peer, every time we connect to a new peer. If the address book has thousands of peers, it can do hundreds of thousands of time system calls every second.

This is unexpected work in sprint 23, but we might need it to make CI more reliable.

Scope

This PR only changes how often Zebra checks the time.
Other changes and cleanups are out of scope.

Solution

Time checks:

• Stop checking the time of every address book peer for each connection attempt
  • Stop sorting all the address book peers for each connection attempt
• Remove redundant peer time checks within the address book
• Stop checking the time 3 times for each address book update
• Stop checking the time multiple times in each address book method

Bug fixes:

• Use saturating time operations to avoid panics

This seems like a bug in Rust or the OS, we really shouldn't be getting earlier times when we remove some time calls. It might also be a multithreading memory consistency issue.

Review

This PR is a blocker for #1873, so we should review it soon.

Reviewer Checklist

• Code implements Specs and Designs
• Existing tests pass

Follow Up Work

This PR makes it more obvious when we're using std::time::Instant instead of tokio::time::Instant. We should fix that in ticket #2258.

[conradoplg]

Looks good, but there seems to be an issue with the usage of BTreeMap

[teor2345]

Bumping this up to medium, because it's required for #1873 in this sprint.

[teor2345]

Looks good, but there seems to be an issue with the usage of BTreeMap

I went with OrderedMap, and added some peer order tests.

[conradoplg]

Looks good!