fix(backup): avoid hang when closing unbound channel

[vividctrlalt]

What

BackupServer.close() can hang if it races start()'s bind, and the keep-alive task can write() after the UDP channel is already closed (logger.error("Exception in send keep alive")).

Why

channel was a plain field next to a volatile boolean shutdown. Those two writes form a Dekker pair: close() must observe a completed bind, and start() must observe shutdown after publishing channel. Without volatile on channel, close() can see a still-null channel, skip channel.close(), and leave closeFuture().sync() blocked.

The keep-alive scheduler is a second window. If the channel is closed while the scheduled task is still eligible to run, MessageHandler.accept() writes to a closed channel. BackupManager.stop() waits for the executor and cannot be used for that first step.

Changes

• Mark BackupServer.channel volatile (Dekker with shutdown).
• After bind in start(), if shutdown is already set, close the channel immediately.
• close() order: stop the keep-alive scheduler without waiting (BackupManager.stopScheduler() — shutdown() only), close the channel, then backupManager.stop().
• BackupServerTest polls isBound() instead of Thread.sleep(1000).
• chooseRandomPort / checkPortAvailable also probe UDP. setReuseAddress after bind is removed (the socket is already bound; the call does not affect the bind that just succeeded and can report a port as free when it is not).

Test plan

• ./gradlew :framework:test --tests org.tron.common.backup.BackupServerTest