Update register stream timestamp more often

### What changes are proposed in this pull request?

During a worker registration stream, we identify stale stream and
`DefaultBlockMaster$WorkerRegisterStreamGCExecutor` recycles them based
on the last-message-received timestamp. This change updates the
timestamp more often.

### Why are the changes needed?

There will be fewer false positives for stale streams. The false
positives are expensive.

### Does this PR introduce any user facing changes?

NA

pr-link: Alluxio#14566
change-id: cid-04a8cc61d7a00933949aed40b2903cdc244e7d64

core/server/master/src/main/java/alluxio/master/block/RegisterStreamObserver.java

@@ -94,6 +94,9 @@ public RegisterWorkerPResponse call() throws Exception {
                 + "Probably %s was exceeded!",
             PropertyKey.MASTER_WORKER_REGISTER_STREAM_RESPONSE_TIMEOUT.toString());
 
+        // Update the TS before and after processing the request, so that when a message
+        // takes long to process, the stream does not time out.
+        mContext.updateTs();
         mBlockMaster.workerRegisterStream(mContext, chunk, isHead);
         mContext.updateTs();
         // Return an ACK to the worker so it sends the next batch
@@ -151,6 +154,9 @@ public RegisterWorkerPResponse call() throws Exception {
                 + "Probably %s was exceeded!",
             PropertyKey.MASTER_WORKER_REGISTER_STREAM_RESPONSE_TIMEOUT.toString());
 
+        // Update the TS before and after processing the request, so that when a message
+        // takes long to process, the stream does not time out.
+        mContext.updateTs();
         mBlockMaster.workerRegisterFinish(mContext);
         mContext.updateTs();