[SPARK-34361][K8S][FOLLOWUP] Fix potentially flaky unit test with multiple resource profiles in ExecutorPodsAllocatorSuite

[attilapiros]

What changes were proposed in this pull request?

Fixing the potentially flaky test in ExecutorPodsAllocatorSuite where multiple resource profiles are used:

• SPARK-34361: scheduler backend known pods with multiple resource profiles at downscaling
• SPARK-33288: multiple resource profiles

Why are the changes needed?

Resource profiles in executor PODs allocator are processed in a nondeterministic order.

Even as the tests are using Maps created by a constructor with two entries and the implementation behind is Map2 where the iterator keeps the ordering in the setTotalExpectedExecutors() method the entries are added to a ConcurrentHashMap totalExpectedExecutorsPerResourceProfileId.

So the resource allocation order might change from run to run.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

Existing unit test.

[attilapiros]

There is one more place to fix

[SparkQA]

Test build #135703 has finished for PR 31719 at commit b291d22.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[attilapiros]

I wanted to be 100% sure that there is no more flakiness in the test. So temporary I have done a small code change and the sorted the map to test both asc and desc ordering. With this temporary code I have found a bug in an earlier test (in SPARK-33288: multiple resource profiles) which was there even before my SPARK-34361 PR.

So to be on the safe side I decided to modify the production code to have a deterministic ordering for allocating the resources( in case of multiple resource profiles).

[SparkQA]

Test build #135707 has finished for PR 31719 at commit 711ea68.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[attilapiros]

closing this as flakiness is not reported by others