fix #4680 : Custom key deserialiser registered for Object.class is ignored on nested JSON

[JooHyukKim]

fixes #4680

Still have below three tests failing.

[ERROR] Errors: 
[ERROR]   DeepNestingUntypedDeserTest.testFormerlyTooDeepUntypedWithArray » StackOverflow
[ERROR]   DeepNestingUntypedDeserTest.testFormerlyTooDeepUntypedWithObject » StackOverflow
[ERROR]   TestAbstractTypes.testDefaultingRecursive:137 » JsonMapping (was java.lang.NullPointerException) (through reference chain: java.util.TreeMap["x"])

[JooHyukKim]

@cowtowncoder This is what I came up with atm. I was hoping maybe you might think of better approach than this? 😅

First, null/non-null check for keyDeser is done at line 228 and line 237
Second, declaring local variable untyped seems a bit unnatural.

[cowtowncoder]

Ok, we might be able to make this work. But one quick idea: how about adding failing unit test first, in a separate PR? I could check that in first. And then hopefully help figure out what to do for actual problem.

FWTW, I suspect this will need to go in 2.19. But we can start PR against 2.18.

[cowtowncoder]

I think this might need to see if we get the "default" key deserializer -- I forget how exactly it is done, maybe regular MapDeserializer has an example -- and if so, ignore it (re-set to null).

(I may be wrong wrt above, but basically wondering if ctxt.findKeyDeserializer() ever returns null or not)

[JooHyukKim]

Thanks! Added commit according to your suggestio and we are down to one last case 👌🏼👌🏼

[ERROR] Errors: 
[ERROR]   UntypedDeserializationTest.testUntypedWithCustomScalarDesers:263 » NullPointer

[JooHyukKim]

Ah, simple mistake, didn't check if we had _customKeyDeserializer not null

[JooHyukKim]

how about adding failing unit test first, in a separate PR?

Makes sense! Made #4685 against 2.18 (so that we can verify behavior since 2.18) @cowtowncoder

[JooHyukKim]

Marked as ready for review and waiting for 2.19 branch out :))