fix(java-client): guard against NPE in FutureGroup.waitAllComplete when future cause is null

[cauchy1988]

Background

Closes #2152. When using batch APIs (Batch.asyncCommit → FutureGroup.waitAllComplete, a.k.a. commitWaitAllComplete), a NullPointerException can occur intermittently.

Root cause

FutureGroup.waitAllComplete builds the failure message by dereferencing fu.cause():

} else {
    results.add(Pair.of(
        new PException("async task #[" + i + "] await failed: " + fu.cause().getMessage()),
        null));
}

A netty Future.cause() returns null not only on success, but also while the future is still incomplete. If await times out before the future ever resolves, isSuccess() is false (so we enter the else branch) but cause() is still null — thus cause().getMessage() throws NPE, crashing the entire batch commit instead of reporting just the single failed task.

The sibling method waitAllCompleteOrOneFail is unaffected because it passes fu.cause() directly to the PException constructor, which tolerates null.

Fix

Guard the dereference and fall back to a placeholder message when cause() is null:

Throwable cause = fu.cause();
String causeMsg = cause != null ? cause.getMessage() : "unknown cause (future not completed)";
results.add(Pair.of(new PException("async task #[" + i + "] await failed: " + causeMsg), null));

This matches the fix suggested by the issue reporter.

Test

Added TestFutureGroup.testWaitAllCompleteHandlesNullCause, which uses Mockito to mock an incomplete Future (await→false, isSuccess→false, cause→null) to reproduce the scenario deterministically, without relying on timing.

Verified both directions:

• With the fix: all 3 tests in TestFutureGroup pass.
• Without the fix (temporarily reverted): the new test fails with the exact NPE reported in the issue:

  NullPointerException: Cannot invoke "Throwable.getMessage()" because the return value of "io.netty.util.concurrent.Future.cause()" is null
      at FutureGroup.waitAllComplete(FutureGroup.java:117)