Get rid of powermock

In this comment https://github.com/apache/pulsar/pull/16884#issuecomment-1201512995 I wrote:

I don’t like powermock as it lacks maintenance. It seems all usage of powermock is Whitebox. We can get rid of it with simple reflections. This can be a separate issue, though.

Thus I file this issue to investigate whether we can remove powermock from our testing framework.

WDTY @merlimat @lhotari @codelipenghui @nicoloboschi?

[tisonkun]

@Shoothzj thank you. I’ll consider this approach when protptyping. There’re some cases we try to hijack third-party lib’s field, like eventLoopGroup in bookkeeper code. In those situations, we should still use reflections. (… or patching upstream, or change testing logics)

But yes, we can use multiple alternatives properly in each concrete case.

[tisonkun]

I draft an incompleted patch on #16997.

However, I notice that there’s some code we hack in “set final” field, like:

https://github.com/apache/pulsar/blob/21634edd30bbe33d231ec73e8045d7666a8ad366/pulsar-broker/src/test/java/org/apache/pulsar/client/api/SimpleProducerConsumerTest.java#L1071-L1072

… or even “set static final” field, like:

https://github.com/apache/pulsar/blob/21634edd30bbe33d231ec73e8045d7666a8ad366/pulsar-functions/runtime/src/test/java/org/apache/pulsar/functions/runtime/thread/ThreadRuntimeFactoryTest.java#L111

It seems powermock-reflect is the best library existing to handle these cases since JDK12+ it’s banned changing final field instead you have to use Unsafe. I tend not to maintain such unsafe hacks within Pulsar codebase. And perhaps we should gradually refactor code to avoid depending on such manners.

cc @Shoothzj @eolivelli @nicoloboschi