[HiCache] Merge HiCache event checks to reduce decode overhead (#30511)

Co-authored-by: alphabetc1 <47200617+alphabetc1@users.noreply.github.com>

python/sglang/srt/managers/scheduler.py

@@ -3226,11 +3226,6 @@ class Scheduler(
             batch.batch_is_full = False
             return batch
 
-        # Eagerly release lock_ref on completed write-through nodes so they
-        # become evictable, improving batch scheduling headroom.
-        if self.enable_hierarchical_cache:
-            self.tree_cache.flush_write_through_acks()
-
         # Check if decode out of memory
         if (kv_full_retract_flag := not batch.check_decode_mem()) or (
             TEST_RETRACT and self.forward_ct % TEST_RETRACT_INTERVAL == 0