test(spigot): guard the post-spawn chat path that closed a 26.3 tunneled session

[minekube-ai-engineer]

Source: Discord ask-support thread

Why

SpigotChatSessionPacketFilter sits in the injected local channel's pipeline
(SpigotDataAddon.onInject puts it directly in front of the vanilla network manager) and it handles
post-spawn traffic. A 26.3 packet shape it cannot handle therefore never fails a login: it throws
with the player already in play, the network manager answers an escaping throwable with
lost connection: Internal Exception: <cause>, and that closes the injected channel - so with it the
Connect tunnel, roughly a second after spawn.

0.15.14 resolved ServerboundChatPacket's constructor by hard-coded parameter types while Minecraft
had moved the signature component to Optional<MessageSignature>, so every plain post-spawn chat
packet from a 1.21.5+/26.x client threw there. #170 fixed the resolution (shipped in 0.15.15) and its
test covers the rewrite by calling the private method directly; nothing asserted the pipeline
consequence, which is the part that ends a session.

What this adds

SpigotChatSessionPacketFilterPipelineTest - test-only, no production change: the real filter in a
real Netty pipeline, at its production position
(connect_data_handler -> connect_chat_session_filter -> network manager), with the network-manager
node modelled the way Paper behaves (record the throwable that escapes a handler, then close the
connection). Asserted:

• the post-spawn packet sequence (chat-session update, chat ack, chat) leaves the session open,
  drops the session-only packets, and delivers the chat to the network manager rewritten exactly once
  • for the 26.3 signed and unsigned component shapes and for the pre-1.21 raw shape (26.2 behaviour
    unchanged), plus the signature-absent variant;
• the resolver refuses an unknown component shape instead of guessing one.

Two one-class stubs (ServerboundChatAckPacket, ServerboundChatSessionUpdatePacket) join the
existing net.minecraft... test stubs so the filter's class-name drop path is exercised too.

RED before / GREEN after

Reverting only the filter file to its 0.15.14 content (tests unchanged) reds 5 of 8:

SpigotChatSessionPacketFilterPipelineTest > postSpawnSignedChatFromA26_3ClientKeepsTheTunneledSessionOpen() FAILED
    a post-spawn packet threw out of the chat filter: [java.lang.IllegalArgumentException: argument type mismatch]
SpigotChatSessionPacketFilterPipelineTest > postSpawnUnsignedChatFromA26_3ClientKeepsTheTunneledSessionOpen() FAILED
SpigotChatSessionPacketFilterPipelineTest > thePostSpawnPacketSequenceKeepsTheSessionOpenAndDropsSessionOnlyPackets() FAILED
SpigotChatSessionPacketFilterPipelineTest > anUnresolvableChatShapeIsRefusedInsteadOfGuessed() FAILED
    java.lang.NoSuchMethodException: ...SpigotChatSessionPacketFilter.chatConstructor(java.lang.Class,java.lang.Object,java.lang.Object)
SpigotChatSessionPacketFilterTest > rewritesChatPacketWhenSignatureComponentIsOptional() FAILED
8 tests completed, 5 failed

On the unfixed revision the throwable is recorded by the network-manager node, not by
CommonDataHandler.channelRead's catch (the captured test stderr contains no printStackTrace) -
which is why the player-visible line is Paper's Internal Exception rather than a quiet close. Live
against Paper 26.3 the same hard-coded lookup is a NoSuchMethodException (the raw 5-arg constructor
no longer exists); the dual-shape test stub surfaces it one step later as
IllegalArgumentException: argument type mismatch from the same lookup.

GREEN: ./gradlew build (what CI runs) is BUILD SUCCESSFUL - 444 test cases (core 375, spigot 33,
velocity 32, bungee 4), 0 failures/errors; the new class passes 6/6.

Scope

This is the missing guard for #170's fix, not a new fix, and it is not a claim about the ~15 %
probabilistic 26.3 post-join teardown that the customer reported - that one is still unlocated.