Migrate legacy reflection properties to use MethodHandles, take 2

[stevenschlansker]

I'm pretty sure I figured out the main reason the previous attempt (#5046) did not work - in order to read modules from databind, we still have to call setAccessible(true) which was happening too late due to the virtual method e.g. BeanPropertyWriter.fixAccess happening after constructor finishes.

Work around this by using a lazy memoized Setter class. Also ends up helping to simplify some serialization wonkiness.

Fixes #2083

[stevenschlansker]

Sorry the first attempt did not go so well, but here is try #2. Instead of changing any of the rules around set accessible, defer using the method / constructors / etc until first-use.
It's unfortunate that the fixAccess / setAccessible is done after the constructor finishes...

I've tested this works with jackson-dataformat-xml jackson-jakarta-rs-providers. Working on one test failure so far in jackson-modules-base still.

[stevenschlansker]

Ok, with FasterXML/jackson-modules-base#295, jackson-modules-base builds too :)

[cowtowncoder]

@stevenschlansker I think I will merge this post rc3 (which I hope to release maybe this weekend).
May be able to merge smaller pieces first, will see if some changes could be applied separately.

[cowtowncoder]

Ok made small change (Exception arg to Throwable) into 3.x, merged to this PR.

[stevenschlansker]

Thanks - please let me know if there's any additional changes needed or validation I should perform.

[stevenschlansker]

@cowtowncoder , while working on this, I ran into the BeanPropertyWriter.fixAccess and AnnotatedMember.fixAccess which is unfortunate in that it mutates state after the constructor completes. I am not sure if it is feasible, but it would be nice if we could push that decision up to the constructor - then more state can be final and we could get rid of the lazy memoized setter, for example

[cowtowncoder]

@stevenschlansker Sounds like a good idea at high level. Unfortunately I don't have more context on this (as in, haven't worked in those aspects lately) so say much about feasibility.

I'll try to get #5094 in tonight, to be done with bigger messing up of introspection, to stabilize things.
(#5093 and #5094 are meant to allow reducing introspection but realized that on short term they won't... but lay some foundation for reduction as next steps. Not directly related to your suggestion, just a sidenote on concurrent changes in related area).

[cowtowncoder]

@stevenschlansker Not sure how much (if at all) this helps, but with #5143 I removed java.io.Serializable from SettableBeanProperty -- so I think UnreflectHandleSupplier and SetterHolder no longer need to be JDK serializable either.

What else do we need to help with issue you mentioned wrt BeanPropertyWriter.fixAccess()?

[cowtowncoder]

Almost certainly no longer needed (I removed most implements java.io.Serializable cases in #5143)

[cowtowncoder]

@stevenschlansker Hoping to merge this in right before 3.0.0-rc4 -- anything you want to change before merging? (we can still do follow-ups of course)