Fixes ByteBuddy AbstractMethodError which could occur when running tests from IDE

[JonasKunz]

We have now had for a time the strange behaviour that certain tests would fail when run from the IDE with the following error:

java.lang.AbstractMethodError: Receiver class co.elastic.apm.agent.bci.bytebuddy.MinimumClassFileVersionValidator does not define or inherit an implementation of the resolved method 'abstract net.bytebuddy.jar.asm.ClassVisitor wrap(net.bytebuddy.description.type.TypeDescription, net.bytebuddy.jar.asm.ClassVisitor, net.bytebuddy.implementation.Implementation$Context, net.bytebuddy.pool.TypePool, net.bytebuddy.description.field.FieldList, net.bytebuddy.description.method.MethodList, int, int)' of interface net.bytebuddy.asm.AsmVisitorWrapper.
	at net.bytebuddy.asm.AsmVisitorWrapper$Compound.wrap(AsmVisitorWrapper.java:746) ~[byte-buddy-1.14.8.jar:?]

I've been able to find the root cause of the issue and fix it with this PR.

Our MinimumClassFileVersionValidator implements AsmVisitorWrapper from byte-buddy-dep, which means it uses the unshaded org.objectweb.asm.ClassVisitor as parameter type.
The error message shows that at runtime the AsmVisitorWrapper interface is picked up from byte-buddy instead, which expects the shaded net.bytebuddy.jar.asm.ClassVisitor.

I assume that when the tests are run from maven the classpath order is slightly different, so that byte-buddy-dep wins over byte-buddy. In the IDE the order seems to be the other way around, causing the failure.

I've fixed it by excluding byte-buddy from the dependencies which pull it in transitively: assertJ and mockito.

What does this PR do?

Checklist

• This is something else
  • I have updated CHANGELOG.asciidoc

[jackshirazi]

I think before merging, it's worth doing a manual test using a snapshot and just run through to see all the normal things expected are displayed in the UI

[JonasKunz]

I think before merging, it's worth doing a manual test using a snapshot and just run through to see all the normal things expected are displayed in the UI

The shanges should (in theory) only have altered the testing classpath. However, I agree that it makes still sense to do a manual test.