protocols/kad/behaviour: Remove false assert on connected_peers

[mxinden]

Given the following scenario:

1. Remote peer X connects and is added to connected_peers.
2. Remote peer X opens a Kademlia substream and thus confirms that it supports
   the Kademlia protocol.
3. Remote peer X is added to the routing table as Connected.
4. Remote peer X disconnects and is thus marked as Disconnected in the routing
   table.
5. Remote peer Y connects and is added to connected_peers.
6. Remote peer X re-connects and is added to connected_peers.
7. Remote peer Y opens a Kademlia substream and thus confirms that it supports
   the Kademlia protocol.
8. Remote peer Y is added to the routing table. Given that the bucket is already
   full the call to entry.insert returns kbucket::InsertResult::Pending { disconnected } where disconnected is peer X.

While peer X is in connected_peers it has not yet (re-) confirmed that it
supports the Kademlia routing protocol and thus is still tracked as
Disconnected in the routing table. The debug_assert removed in this pull
request does not capture this scenario.

Fixes #2092

@FelipeRosa @izolyomi would you mind taking a look and confirm that this fixes your issues as well?

[romanb]

Makes sense to me. I guess this was an oversight when implementing the delayed insertion into the routing tables (i.e. only after protocol confirmation).

[izolyomi]

Using this new dependency version 0.39 seems to have a ton of breaking changes compared to 0.38. I'd have to understand all the changes and fix everything to test, I might stick with the old version for the moment.

[FelipeRosa]

Thanks for the work @mxinden and @romanb

Sure, I'll take a look at it later today.

However, I believe that should solve the problem because what I ended up doing to prevent triggering the debug_assert was setting Kademlia's behaviour kbuckets insert to KademliaBucketInserts::Manual and adding DHT peer addresses manually at a later stage (after the peer has sent Identify information and said it supports the kad protocol).

Since I've done that I have not seen the debug_assert being triggered and everything seems to be working fine when using the kad DHT.

[izolyomi]

I've finally upgraded to the unreleased 0.39, it wasn't too hard to fix all breaking changes after all. I can confirm that I haven't encountered debug_assert failing even once.

[mxinden]

Thanks everyone for the input and thanks @romanb for the review!