fix: network manager event loop flush, compression lookup, and shutdown order

[GamingOP69]

Summary

Fixes packet flushing thread safety in NetworkManager, pipeline key lookup during compression setup, sendQueuedPackets error recovery, and clean shutdown ordering in ServerConnection.

Problem & Root Cause

• NetworkManager.flush() called channel.flush() directly from non-I/O threads (such as the main server tick thread). Netty channel pipelines are not thread-safe for direct off-thread pipeline manipulations, which could lead to unflushed packets sitting in channel buffers.
• setupCompression looked up the handler with key "decompress" instead of "compress", failing to insert the compression handler in the expected pipeline position.
• In PlayerConnection.sendQueuedPackets(), if packet serialization threw an exception during queue draining, enableAutomaticFlush() would never get called, permanently stalling the connection's automatic flushing.
• ServerConnection.stopServer() shut down event loop groups before closing the listening channels, throwing NullPointerException and resource leak warnings on server shutdown.
• LazyInitVar lacked double-checked locking, allowing race conditions during concurrent initialization of Netty transport groups.

Solution

• Updated NetworkManager.flush() to execute flushes via channel.eventLoop().execute(...) when called outside the I/O event loop.
• Fixed the compression pipeline key lookup to "compress".
• Wrapped sendQueuedPackets() draining in a try-finally block to guarantee enableAutomaticFlush() runs even if an exception occurs.
• Ensured listening channels are closed before terminating Netty event loop groups in ServerConnection.
• Added double-checked locking to LazyInitVar with LazyInitVarTest concurrency unit test.