chore(patches): revert the "respect max retries" patch

That PR that needs to be reverted introduces a minor behavioral
change: when Nginx retrieves a connection from the keepalive pool and
finds it broken during its use, it will add an additional retry until
it has exhausted all the connections in the keepalive pool and
attempts to establish a new connection with the upstream server in
the next retry round.

In Nginx's perspective, retrying on a broken connection is not
considered a real retry. This is also the default behavior of Kong.

That PR that needs to be reverted modifies this behavior by treating
the retry on a broken connection as a valid retry and, at most,
retrieving 5 (default) connections from the keepalive pool. If all of
them are broken, the balancer will stop retrying, even if there are
still connections available in the keepalive pool. It will also not
attempt to establish a new connection with the upstream server.
Although this is a very rare case, in order to maintain consistency
in behavior, we believe it is necessary to be more cautious.

Therefore, we intend to adopt a difficult but correct fix: addressing
the issue regarding statistics

Signed-off-by: tzssangglass <[email protected]>