[client] fix: throw LogOffsetOutOfRangeException when scan offset ttl log

[gyang94]

Purpose

Linked issue: close #521

Brief change log

Based on comments in issue #521 , this PR did:

1. remove the empty records check in LogFetcher, to pass the CompletedFetch (with error) into logFetcherBuffer

if (!MemoryLogRecords.EMPTY.equals(logRecords)) {

p.s. Based on the comments from @loserwang1024 , this empty records check may be necessary for some cases.

In oder to not signal notEmptyCondition, add completed fetch to buffer until log records is not empty.

I guess it is going to avoid several null records CompletedFetch (without error) in the first several polls.
But we can create a new issue ticket to resolve those cases.

2. throw IllegalStateException when LogOffsetOutOfRangeException occurred in LogFetcherCollector, to show stack trace.

Tests

Add FlussLogITCase.
This test need to set a small ttl and wait for log clean up. It may not be good to reuse the cluster in existing IT cases. Thus add a new IT case for it.

API and Format

Documentation

[wuchong]

Hi @gyang94 do you have time to update the PR? If not available in the recent days, I can help to update it and merge it.

[gyang94]

@wuchong Thanks for review. I will update it in this or next week.

[gyang94]

@wuchong updated. please check.

[wuchong]

#588 has fixed this problem. So I rebased the branch and kept the tests to verify that the problem is fixed. Will merge it once CI passes.