Fix TCP fallback with multiple nameservers [Bug #8285]

[opti]

Originally reported here https://bugs.ruby-lang.org/issues/8285, but the patch has never been finalized.

We stumbled on the same issue under the following conditions:

• Multiple nameservers are configured for Resolv::DNS
• A nameserver falls back from UDP to TCP requester due to the message size
• TCP request hits Resolv::DNS timeout
• Resolv::DNS retries the next nameserver

Expected result:

Resolv::DNS successfully falls back to the next nameserver, using TCP request.

Actual result:

Resolv::DNS::Requester::RequestError: host/port don't match is raised.

This pull request:

• Finalizes original patch proposed by @jmehnle.
• Reuses open TCP connection (by @hanazuki)
• Adds a test case for the scenario above.

[opti]

@sorah, @nobu any chance this can be reviewed?

[opti]

@hanazuki These are valid points. Thank you for the detailed explanation. I've added your commit to this pull request. Thanks for looking into it.

[opti]

@hanazuki any chance this fix can be merged?

[hanazuki]

+1 to merge. I'm not the maintainer and cannot merge or approve PRs.

What do you think, @sorah @nobu?