[Test][C++] Fix flaky ReaderTest.testAsyncRead

[zhanglistar]

Fixes #542

Motivation

ReaderTest.testAsyncRead was flaky because it assumed the i-th readNextAsync callback would receive the i-th message. With partitioned topics or async scheduling, callbacks can complete in any order, so the assertion content == "my-message-" + std::to_string(i) sometimes failed (e.g. expected "my-message-8" but got "my-message-9"). This change makes the test order-independent so it passes for both single- and multi-partition topics.

Modifications

• In testAsyncRead, replaced order-dependent assertions with a set-based check: fire 10 readNextAsync calls, collect the 10 message bodies in a fixed array using std::atomic<int> for the index (no mutex), wait until 10 messages are received, then assert that the set of received strings equals {"my-message-0", ..., "my-message-9"}.
• Removed the assumption that callback completion order matches send order so the test is stable for partitioned and non-partitioned topics.

Verifying this change

• Make sure that the change passes the CI checks.

This change is already covered by existing tests:

• Run: ./pulsar-tests --gtest_filter='*testAsyncRead*'
• Passes for both parameterized cases (single partition and partitioned topic); no flaky failures from callback order.

Documentation

• doc-not-needed
  Test-only change; no user-facing or API documentation updates required.

[zhanglistar]

@BewareMyPower Another flaky test https://github.com/apache/pulsar-client-cpp/actions/runs/22660933680/job/65680996230?pr=545