fix(java): define generated serializers as normal classes

[mandrean]

Summary

Define generated serializers as normal classes instead of hidden nestmates.

The first commit adds a failing regression test for async serializer JIT with an allow-all TypeChecker. The fix updates neighbor-based codegen to use the normal class definition path and adjusts the JDK 25 class-shape assertion.

Closes #3790

Validation

• Java 25 focused JITContextTest
• Java 25 focused CodecUtilsTest
• Java 25 focused ObjectCodecBuilderTest
• Java 17 focused JITContextTest
• spotless:check
• checkstyle:check

[chaokunyang]

Could you check whether fory 1.3.0 still have this issue? I released 1.3.0 today

[chaokunyang]

And I think we should not define generated serializer as hidden class, define it as normal class is enough. I define it as hidden class to access private fields directly, but janino can't compile it. With #3778, field access on jdk25 is not a performance issue anymore.

[mandrean]

Could you check whether fory 1.3.0 still have this issue? I released 1.3.0 today

1.3.0 actually fails to deserialize some of our 1.2.0 binary blobs altogether... Seems like there is a binary incompatible breaking change in 1.3.0? I'll see if I can come up with a minimal reproducible for that, and create a separate issue.

I moved back the PR to draft, I'll refine it a bit before turning it ready for review again