fix: Address race condition in generation handling after sealDir

Move the generation load to immediately after sealDir() returns but
before reacquiring dh.mu. This ensures we capture the generation at
the point when sealDir completes, avoiding a race where another thread
could call sealDir() between our unlock and the generation load.

The sequence is now:
1. Unlock dh.mu
2. Call sealDir()
3. Load generation (while dh.mu is still unlocked)
4. Reacquire dh.mu
5. Update dh.generation

This ensures we get the most accurate generation value while still
avoiding the deadlock.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>

Author: ovaistariq

core/dir.go

@@ -636,9 +636,11 @@ func (dh *DirHandle) listObjectsFlat() (start string, err error) {
 		// We must release dh.mu before calling sealDir to avoid deadlock
 		dh.mu.Unlock()
 		dh.inode.sealDir()
+		// Reload generation immediately after sealDir completes to get accurate state
+		currentGen := atomic.LoadUint64(&dh.inode.dir.generation)
 		dh.mu.Lock()
 		// Update our generation to match the new state
-		dh.generation = atomic.LoadUint64(&dh.inode.dir.generation)
+		dh.generation = currentGen
 	}
 
 	dh.inode.mu.Unlock()