why apply NewHeader，s.KV().Rev() ，not use SharedBufReadTxMode？ #20184
Description
What would you like to be added?
During stress testing of lease granting operations on an NVMe disk, I observed a maximum QPS of only around 25,000. The bottleneck appears to be caused by s.KV().Rev() using the ConcurrentReadTx mode. This mode requires copying the buffer memory each time the revision is accessed.
Given that this operation (s.KV().Rev()) only needs to retrieve the current revision number (a single version ID), why can't the SharedBufReadTx mode be used instead? This seems like it would avoid the costly memory copy operation and significantly improve performance under high concurrent lease grant loads.
Why is this needed?
significantly improve performance under high concurrent lease grant loads.