Preserve defaultValue literals

[leebyron]

Depends on #3092
Fixes #3051

This change solves the problem of default values defined via SDL not always resolving correctly through introspection by preserving the original GraphQL literal in the schema definition. This changes argument and input field definitions defaultValue field from just the "value" to a new GraphQLDefaultValueUsage type which contains either or both "value" and "literal" fields.

Since either of these fields may be set, new functions for resolving a value or literal from either have been added - getLiteralDefaultValue and getCoercedDefaultValue - these replace uses that either assumed a set value or were manually converting a value back to a literal.

Here is the flow for how a default value defined in an SDL would be converted into a functional schema and back to an SDL:

Before this change:

(SDL) --parse-> (AST) --coerceInputLiteral--> (defaultValue config) --valueToAST--> (AST) --print --> (SDL)

coerceInputLiteral performs coercion which is a one-way function, and valueToAST is unsafe and set to be deprecated in #3049.

After this change:

(SDL) --parse-> (defaultValue literal config) --print --> (SDL)

[andimarek]

I think that is the right direction and as far as I understand you want that merged before the other bigger changes.

The more comprehensive end state we are aiming for In GraphQL Java is this:

public enum ValueState {
    NOT_SET,
    LITERAL,
    EXTERNAL_VALUE,
    INTERNAL_VALUE // this is deprecated and should not be used going forward, will be removed in the future
}

to capture the different states of the default value:
NOT_SET you probably don't need because you JS has undefined.

This is needed because we need to differentiate between external input value which will be converted to literal via valueToLiteral and the legacy behaviour of using a not well defined already coerced internal value, which will continue to use astFromValue.

[leebyron]

Yeah, I'm treating this as a prerequisite dependency of #3049, similar to a few of the other changes I have up right now. I'm hoping to keep that diff as focused as possible, and this change should have very little effect other than resolving this issue.

The more comprehensive end state we are aiming for In GraphQL Java is this

That sounds right to me! Yes we use undefined to represent not set. I'm also considering performing the uncoerce step at schema construction time so that "INTERNAL_VALUE" is only a state the configuration can be in, rather than the final reified schema types. But that's for the later change :)

[leebyron]

Updated to do proper memoization and error if multiple sources of truth are presented in a config

[yaacovCR]

Merged as #3810