Fix possible regression issue caused by ProposeWaitEmptyBlockTimer

Possible scenario:
* Schedule timeout for `ProposeWaitEmptyBlockTimer` on height: n
* On timeout, timer loop queues a call to `Tendermint::on_timeout` on the
timer worker
* Blocks are imported at the same time with the timeout.
  * Block importer holds lock for Tendermint, calls `new_blocks`
* The worker checks whether it is cancelled, but it is not cancelled yet.
  * The worker waits for the lock.
* several `move_to_height`, `move_to_step` are called by new_blocks
  * height is not n anymore.
  * `move_to_step` clears timer, but the check is already passed and the
worker waits for the lock.
  * It finally releases the lock.
* The waiting worker calls Tendermint::on_timeout and the timeout for
`ProposeWaitEmptyBlockTimer` calls `move_to_step(Prevote)`
  * The timeout was set for the height n, but the code in the timeout
reads changed height.

Author: foriequal0

core/src/consensus/tendermint/worker.rs

@@ -1232,8 +1232,15 @@ impl Worker {
                 TendermintState::ProposeWaitEmptyBlockTimer {
                     block,
                 } => {
-                    cdebug!(ENGINE, "Empty proposal timer is finished, go to the prevote step and broadcast the block");
-                    self.submit_proposal_block(block.as_ref());
+                    if self.height == block.header().number() {
+                        cdebug!(
+                            ENGINE,
+                            "Empty proposal timer is finished, go to the prevote step and broadcast the block"
+                        );
+                        self.submit_proposal_block(block.as_ref());
+                    } else {
+                        cwarn!(ENGINE, "Empty proposal timer was for previous height.");
+                    }
                 }
                 _ => {
                     cwarn!(ENGINE, "Empty proposal timer was not cleared.");