Add cluster_name property to avoid several clusters merging accidentally

[xvello]

As described in quickwit-oss/quickwit#1260, we need a mechanism to ensure clusters do not accidentally merge when IPs are reused.

This PR:

• introduces a required cluster_name argument when setting up the cluster
• adds that cluster name to outgoing Syn messages
• adds a BadCluster message type
• if the cluster name in the Syn message does not match the peer's cluster name, a BadCluster message is returned instead of a SynAck

Additional notes

• This PR changes the wire format, but I did not see any versioning mechanism in the protocol. We should merge this before v0.3 is released, to sidestep any compatibility issue.

[fulmicoton]

As you suggested in the PR description, could you add an error message to the protocol?

As you mentioned it is nicer for testing, but also,I suspect a common problem that will occur is a misconfigured typo in the conf of one of the server. The user will likely investigate the log of the server that cannot connect to the cluster first and an explicit error message there might be nice.

[fulmicoton]

I create #37 and #38 too.

[xvello]

I just pushed a commit introducing a BadCluster message, instead of silently dropping the message.

I started writing the code with a more generic Rst message, but I could not get a relevant error displayed on the misconfigured node without extending it with a RstReason enum, which added unneeded complexity.

[xvello]

ignore_oversized_payload is failing, but I think this test has lost its initial meaning and is not useful anymore. Discussing it on discord.

[fulmicoton]

@xvello FYI: in tracing ?val will use Debug. %val will call Display.
For a String you want %val

Alternatively, you can also try to get a &str and avoid any modifier. key=val.as_str()

[xvello]

Thanks for the review.
I'm opening a PR to quickwit to introduce the new config option.