refactor: extract kbuckets into separate structure

[morph-dev]

What was wrong?

The kbuckets are widely used and they are wrapped in Arc<RwLock<>> for asynchronous access.

This can be dangerous combination as it's easy to hold the kbuckets lock and try to acquire some other lock or do .await. This was the root problem of the #1390 .

How was it fixed?

I wrapped Arc<RwLock<KBucketsTable<NodeId, Node>>> in a structure (that I named SharedKBucketsTable).
Every time lock is acquired, it's held only for the minimum amount of time, during one function call.

This approached is recommended in https://draft.ryhl.io/blog/shared-mutable-state/ , which is referenced from: https://tokio.rs/tokio/tutorial/shared-state#restructure-your-code-to-not-hold-the-lock-across-an-await

I tried to minimize functionality changes. Here are highlights:

• interested_enrs and batch_interested_enrs are just moved from gossip.rs
• insert_or_update_discovered_nodes - re-implementation of previous OverlayService::process_discovered_enrs
  • original implementation was handling a case that is not possible so I removed it (for reference, inserted nodes can't be inserted as pending because they are inserted as disconnected)
• nodes_by_distance - original implementation wasn't working correctly
  • it would call KBucketsTable::nodes_by_distances but it would potentially later filter entries, resulting in returning less results than desired

Note: While this PR is big, I would say that at least 2/3 are test related changes. If desired, I can try to split it up into smaller PRs for easier review.

To-Do

• Clean up commit history and use conventional commits.

[KolbyML]

looks good I didn't see anything wrong when reviewing. I also ran the PR through hive and all tests pass there as well.