Remove TcpChannel#setSoLinger method

[Tim-Brooks]

This commit removes the dedicated setSoLinger method. This simplifies
the TcpChannel interface. This method has very little effect as the
SO_LINGER is not set prior to the channels being closed in the abstract
transport test case. We still will set SO_LINGER on the
MockNioTransport. However we can do this manually.

[elasticmachine]

Pinging @elastic/es-distributed

[original-brownbear]

LGTM

[s1monw]

@tbrooks8 @original-brownbear I am not sure I follow. Didn't we add this to make some tests be more reliable? Now we only set this on the the MockNioTransport. Should we also set it manually on all the others?

[original-brownbear]

@s1monw

turning off the linger has outlived its usefulness due to other changes. When I first added this a year ago it was applied to all channels before closing them which had a massive impact. This caused some trouble so the logic here https://github.com/elastic/elasticsearch/pull/35924/files#diff-3b8055db0941b6b23392b71384b69678L353 started checking for the life-cycle state so that we only RST from stopped TCP transports. This makes sense, but as @tbrooks8 points out is very rarely the case (we generally close first then stop).
=> this logic doesn't actually set linger a lot
Also, with @tbrooks8 lowering the connection count in tests a lot via some recent changes this isn't practically necessary any longer as far as I can see.

Should we also set it manually on all the others?

I don't think we should add this to the production ones. In practice their tests' (lingering) connection counts are low enough now to not cause trouble. We could maybe set linger to 0 for the blocking mock network implementation as well, but practically it doesn't seem necessary at this point. Also, there has been some recent discussion to maybe remove the blocking mock network altogether since its kind of redundant. See #34458 (comment)

[s1monw]

sounds good to me @original-brownbear thanks