fix(exit): flush 0x03 Redirect before WorldState cleanup races the socket

[Caeldeth]

Summary

On PacketHandler_0x0B (endSignal=0), the 0x03 Redirect packet was queued after Map.Remove(user) had already enqueued a 0x0E AoiDeparture, and with the default transmitDelay = 1200ms. Two compounding problems:

1. FlushSendBuffer breaks out when it hits a delayed packet after buffering a no-delay one — so the 0x03 was always deferred to the next 50ms ProcessOutbound tick rather than going out in the immediate flush that Client.Redirect's Enqueue(flush: true) was trying to force.
2. When the deferred 0x03 did get picked up, its 1200ms Task.Delay easily outlived the live-socket window. Eager-closing clients (or downstream socket teardown from the cleanup below this line) meant the Redirect never reached the wire.

The visible client symptom: a spurious "Connection Lost" popup on Exit Game because the client interpreted the silent socket close as an unexpected disconnect rather than a logout-and-redirect.

Fix

Two changes in one block (World.cs):

• Reorder so SendRedirect runs first. Its flush: true Enqueue triggers an immediate FlushSendBuffer against an empty queue — the 0x03 is the only packet, so the "break on delayed packet after no-delay" rule doesn't fire.
• Drop transmitDelay from 1200 to 50 so the inter-packet hold is short enough that the live-socket window comfortably covers it.

user.SendRedirect(this, Game.Login, user.Name, false, transmitDelay: 50);
user.UpdateLogoffTime();
user.Map.Remove(user);
// ... rest of cleanup

Diagnosis trail

Verified via client-side packet log (Chaos.Client notice-debug.log). Before fix:

outbound 0x0B [0]   ← client confirm
inbound  0x0E       ← AoiDeparture
World -> Disconnected   (~50-100ms after confirm)

No 0x03 ever arrives at the client. After this fix the 0x03 should appear ahead of the disconnect and the client's normal redirect flow takes over.

Test plan

• Smoke test: log in to world, click Exit Game in client, verify clean return to login (no "Connection Lost")
• Verify other Redirect paths still work (heartbeat-fail kicks, character-switch via Logoff, login→world transition)
• No regression to the endSignal=1 (query) branch

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 31.84%. Comparing base (d4e7e52) to head (2b24087).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #335      +/-   ##
==========================================
+ Coverage   31.79%   31.84%   +0.04%     
==========================================
  Files         339      339              
  Lines       26485    26485              
  Branches     3681     3681              
==========================================
+ Hits         8422     8433      +11     
+ Misses      17349    17337      -12     
- Partials      714      715       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.