Add KeyDeserializer to WireSafeEnum

[jhaber]

So that Jackson can deserialize maps with WireSafeEnum as keys (which has nicer ergonomics and reduces friction to using WireSafeEnum, since plain enums support this already).

This test is probably most illustrative of the new functionality:
464f920#diff-8095941f2209daf7eb627be2a5d6be0953a3073a8a4e4e91d239abbe945fcadbR537-R549

I thought we might need a KeySerializer as well, but it doesn't seem to be necessary.

Unfortunately this PR requires upgrading Jackson in order to pick up this feature: FasterXML/jackson-databind#2503

We're already using 2.12.6 internally, so this doesn't actually represent a change for us. However upgrading Jackson caused a few test failures that I need to investigate.

@stevie400 @Xcelled @suruuK @kmclarnon

[jhaber]

Unfortunately JsonDeserializer and KeyDeserializer are both abstract classes, so we can't implement both in a single class. This required a bit of refactoring to extract some shared methods from the existing deserializer class, which makes the diff look a lot bigger than it actually is. I also refactored the tests so that all of the json reading/writing tests will cover plain objects and maps at the same time (to get better test coverage without having to duplicate all the tests).

I did this over 3 commits to make it a bit easier to review. First commit just moved some methods out of the inner class to the top level class:
2e0e3fa

Second commit was just test refactoring so that we're asserting on a stream of objects rather than a single object:
1efdd24

Third commit was actually adding the KeyDeserializer, wiring it into the tests, and upgrading Jackson so that the tests pass:
464f920

[jhaber]

Tests fixed in 6fcb8dc, they just needed to be updated to handle some minor changes to exception messages in the new Jackson version.

[jhaber]

Some of these version overrides can go away once we upgrade basepom to 25.8. I tried doing that in this PR, but it automatically enabled prettier, which I didn't want to do as part of this PR (it reformatted all the code)

[kmclarnon]

LGTM 👍

[jhaber]

This should be pretty safe because deserialize to a map key would've just failed before. And if anyone was registering a custom key deserializer, that would still take precedence.