HDDS-15328. Fix sleep time in RetryInvocationHandler

[adoroszlai]

What changes were proposed in this pull request?

Sleep period in RetryInvocationHandler should be waitTime, not delay (see #10310 (comment)).

1. RetryInfo.retryTime is calculated when it is created:
   

   ozone/hadoop-hdds/common/src/main/java/org/apache/hadoop/io_/retry/RetryInvocationHandler.java

   Lines 253 to 254 in c917a6e

   	this.delay = delay;
   	this.retryTime = Time.monotonicNow() + delay;

2. waitTime is calculated before sleep as:
   

   ozone/hadoop-hdds/common/src/main/java/org/apache/hadoop/io_/retry/RetryInvocationHandler.java

   Lines 85 to 86 in c917a6e

   	synchronized Long getWaitTime(final long now) {
   	return retryInfo == null? null: retryInfo.retryTime - now;

3. but then original RetryInfo.delay is used for actual sleep time:
   

   ozone/hadoop-hdds/common/src/main/java/org/apache/hadoop/io_/retry/RetryInvocationHandler.java

   Lines 131 to 136 in c917a6e

   	final Long waitTime = getWaitTime(Time.monotonicNow());
   	LOG.trace("#{} processRetryInfo: retryInfo={}, waitTime={}",
   	callId, retryInfo, waitTime);
   	if (waitTime != null && waitTime > 0) {
   	try {
   	Thread.sleep(retryInfo.delay);

Since this is such a small change, also make some improvements in this area:

• Remove some leftover references to async call handling
• Port TestRetryProxy from Hadoop
• Change test-client profile in pom.xml to include tests ported from Hadoop (more tests to come in follow-up PRs)

https://issues.apache.org/jira/browse/HDDS-15328

How was this patch tested?

Tests run: 10, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 2.274 s -- in org.apache.hadoop.io_.retry.TestRetryProxy

https://github.com/adoroszlai/ozone/actions/runs/26150721145

[ivandika3]

@adoroszlai Thanks for catching this issue, LGTM +1.

I think this will subtly change the retry wait behavior where instead of waiting for a fixed delay, it will wait based on the calculated time when RetryInfo is created. However, I think the difference should not be noticeable in most cases since RetryInfo creation in handleException and waitTime calculation in processWaitTimeAndRetryInfo should be similar most of the time (unless there is a huge pause between handleException and processWaitTimeAndRetryInfo or other cases like when retryTime is not set to null in processRetryInfo somehow.