fix(fake-death): fix client death-restart sync (stuck Game Over, stale corpse, blocked doors)

[lisniuse]

Problem

In co-op, when both players go down and the host restarts the run (all-players-downed), the client could end up:

• Stuck on the Game Over screen at 1 HP — never respawning or healing, sometimes crashing with Null access .curCine.
• After the host's restart, the host kept seeing the client as a corpse pinned at the old death position, and the client could not interact with doors (exit doors gated as if still downed).

Reproduction

1. Two players in co-op.
2. Client goes down first, then the host goes down (all players downed → host restarts the run).
3. Observe the client.

Root cause

1. In-place level reload pre-empts the full restart (stuck Game Over / crash).
On the all-downed restart the host broadcasts a new run seed + the new level graph. On the client:

• the seed triggers QueueClientRestartFromHostSeed → a queued full launchGame restart;
• the level graph then triggers TryTriggerLevelGraphReload → an in-place reloadAfterBossRuneModif, which runs first and reloads the old run in place, swallowing the full restart. The client is left in the old run at 1 HP with Game Over; a bad curCine state can crash with Null access .curCine.

2. Revived state never broadcast (stale corpse + blocked doors).
ResetDownedPlayersForRestart called ResetFakeDeathState(sendNetworkUpState: false), so the restarting client never told the peer it was no longer downed. The host's _remoteDowned[client] stayed set, so ReceiveGhostCoords pinned the client ghost at the corpse position and LevelExitSync gated its interactions.

Fix

• LevelSync: guard TryTriggerLevelGraphReload / TryTriggerBossRuneReload with ShouldSuppressClientLevelReload() — skip the in-place reload while the local player is downed or a full restart is pending, so the queued launchGame restart wins. Map consistency is unaffected (still synced via generateGraph + TryApplyRemoteLevelGraph).
• GameMenu: add a client-restart-pending flag (set when QueueClientRestartFromHostSeed queues, cleared on MarkInRun) to cover the window where _localFakeDead is already cleared but the new run hasn't started yet.
• FakeDeath: ResetDownedPlayersForRestart now broadcasts the not-downed state (sendNetworkUpState: true) so the peer clears stale remote-downed tracking.

Notes

• Trade-off: while a full restart is pending, the client skips the in-place hot-reload of the current level; the host changing the boss rune mid-level is reflected on next level entry instead of instantly. Normal forward progression is unaffected.
• Verified in a live 2-player session (client-first death → host death → restart): client now respawns at the start level at full HP, the host sees the client move normally (no corpse), and door interactions work.

🤖 Generated with Claude Code