Improve quality of LST: replace applicable markers with LST elements

[traceyyoshima]

Replace:

• AnnotationUseSite:
  The J tree expects J.Annotation(s), a final class. In K, there is an additional suffix after the use site annotation like @file : .... This change may require an update to J.Annotation since various classes require it.
• IsNullable:
  Nullable is a generally applicable concept across J and every implementation of J, and may be generally fixable. Deprecate IsNullable marker and use K.Unary instead #489
  Deprecate IsNullable marker and use J.NullableType instead #497
• IsNullSafe:
  Possible solution: Add an IsNullSafe Expression pass-through to contain the applicable expression.
• Modifier: The modifier marker was supposed to be replaced in the PSI parser with the J.Modifier#keyword, and appears to still be in use.
• TypeReferencePrefix
  Possible solution: Add a TypeReference TypeTree tree to contain the applicable expression.

Update:

• Elvis:
  This marker exists in both Groovy and Kotlin, and will likely be reused elsewhere. We should probably add this to rewrite-java and use the reuse the marker.

[kunli2]

For AnnotationUseSite marker, I propose we create a new model K.Annotation to present KtAnnotation

[kunli2]

Elvis marker is not used anymore in Kotlin and has been changed a K.Binary, so we can deprecate it directly here.

[traceyyoshima]

• The J tree expects J.Annotation(s), a final class. In K, there is an additional suffix after the use site annotation like @file : .... This change may require an update to J.Annotation since various classes require it.

How does this work for the requirements of J.Annotation?

[kunli2]

I double checked IsNullSafe marker, there are total 3 usage scenarios

1. J.FieldAccess, example: test ?. property, the space before ?. is handled by leftPadded
2. J.MethodInvocation, example:

              val filter = null   ?. let { _ ->
                  { false
                  }
              }

the space before ?. is handled by suffix of the select (RightPadded)

3. J.TypeCast, example val map = mapOf(Pair("one", 1)) as? Map<*, *>, the space before ?. is handled by the prefix of J.TypeCast.clazz

so I think the prefix in this marker is not necessary, we can just deprecate the prefix field and keep the marker there.
@knutwannheden @traceyyoshima what do you think?

[knutwannheden]

Let's also try to be careful to not break LST deserialization as we do these refactorings. In moderne-ast-write we can set up regression tests, so that we know if we need to customize deserialization or not.