test(pubsub): fix flakiness and formatting issues in schemas acceptance tests

[torreypayne]

closes: #34025

This PR fixes flakiness and formatting issues in the schemas_test.rb acceptance tests for google-cloud-pubsub samples. These issues were causing CI failures, particularly on newer Ruby versions (3.3+).

Root Causes & Fixes

1. Ruby 3.3 Hash Formatting Difference:

Issue: Avro subscriber tests used assert_output with exact string matching. The expected output had spaces around => (e.g., {"name" => "Alaska"}), but Ruby 3.3 represents Hashes without spaces (e.g., {"name"=>"Alaska"}), causing consistent failures on attempt 1.
Fix: Updated the assertions to use space-tolerant regular expressions (e.g., /Received a binary-encoded message:\n{"name"\s*=>\s*"Alaska",\s*"post_abbr"\s*=>\s*"AK"}/).

2. Subscription Propagation Delay (Flakiness):

Issue: Messages were being published immediately after subscription creation. Because Pub/Sub subscription creation is eventually consistent, this occasionally led to messages being lost if they were published before the subscription was active in all zones.
Fix: Added sleep 5 after subscription creation in the Avro tests, matching the behavior already present in the Proto tests.

3. Out-of-Order Message Delivery:

Issue: The revisions test published two messages (one for Rev A, one for Rev B) and asserted their receipt using a single regex that enforced order (Alaska then California). Since Pub/Sub delivery is unordered by default, they could arrive in reverse order, causing the assertion to fail.
Fix: Changed the assertion to use capture_io and separate assert_match calls for each expected message, making the test order-independent.

4. Shadowed Errors in expect_with_retry:

Note: The test helper expect_with_retry retries the test block up to 5 times. If the first attempt failed (due to the issues above) but acknowledged the messages, subsequent attempts would receive nothing and fail with actual: "". Minitest then reported the last failure, shadowing the actual root cause of the initial failure. Fixing the root causes above prevents this retry-failure loop.

Verification Results
Verified locally on fix-acceptance-tests-flakiness branch (which is based on main and only contains these fixes):

GOOGLE_APPLICATION_CREDENTIALS=... GOOGLE_CLOUD_PROJECT=helical-zone-771 bundle exec ruby acceptance/schemas_test.rb

Output:

Finished in 392.020107s, 0.0332 runs/s, 0.1377 assertions/s.
13 runs, 54 assertions, 0 failures, 0 errors, 0 skips
All existing tests passed successfully on the first attempt.

[quartzmo]

I wonder if instead of using a hard-coded sleep (guaranteed to add 5s to the test suite), we could use expect_with_retry from google-cloud-ruby/google-cloud-pubsub/samples/acceptance/helper.rb instead? It's used in other tests for this gem. Maybe something like:

# pubsub_subscribe_avro_records
expect_with_retry "pubsub_subscribe_avro_records" do
  # Publish the message inside the retry loop
  publisher = pubsub.publisher @topic.name
  publisher.publish record.to_json

  expected = /.../
  assert_output expected do
    subscribe_avro_records subscription_id: @subscription.name, avsc_file: nil
  end
end

[torreypayne]

ya I don't like the sleep 5 solution at all and I've been digging into an alternative fix; let me try that out.