[vindeckyy]

Fixes #4309.

@ValidHost and TcpSocketManagerFactory both treat a host that does not resolve now as invalid, so the appender is discarded at configuration time. A host that only appears in DNS later (a Kubernetes Service created after the JVM starts) never recovers without a restart, and reconnectionDelayMillis cannot help because no appender or manager is left.

Changes

• ValidHostValidator still rejects null, empty, malformed names and out-of-range IPv4 literals, but accepts a syntactically valid host name that does not resolve yet and logs a warning. The name is resolved when connecting.
• TcpSocketManagerFactory no longer returns null on UnknownHostException when a reconnection delay is configured; it creates a manager with no socket and lets the existing Reconnector resolve the host on every attempt. Appenders always get a delay (getSocketManager defaults it to 30s), so this also covers default configurations.
• DatagramSocketManager/DatagramOutputStream resolve the host on the first datagram instead of failing construction, and retry while the host stays unresolved.
• AbstractSocketManager#getContentFormat falls back to the host name while no address is known.
• Updated the @ValidHost Javadoc and the plugin manual.

Tests

SocketAppenderReconnectTest#unresolvable_host_should_be_resolved_on_reconnect fails before the change (appender dropped for an unresolvable host) and passes after it. ValidHostValidatorTest covers the accepted and rejected names, and DatagramOutputStreamTest covers the UDP retry path.

Targeted run: 82 tests in SocketAppender*Test, SyslogAppender*Test, Tls*Test, SslSocketManagerTest, ValidHostValidatorTest, DatagramOutputStreamTest, 0 failures. spotless:check passes.