Replace BiMap with plain maps #861

graphql.kickstart.tools.util.BiMap is a copy of Guava's BiMap we added when we dropped Guava (661769e). Neither Java nor Kotlin has a bidirectional map, so it stuck around. After #851 it doesn't really pull its weight anymore.

Where it's used:

• SchemaClassScanner builds a BiMap<TypeDefinition, JavaType> (TypeClassDictionary). SchemaParser reads it both ways: dictionary[definition] for enum classes and dictionary.inverse() for InterfaceTypeResolver and UnionTypeResolver.
• BiMap#put throwing value already present is the only thing that enforces one class per output type. It shows up as Error creating bimap of type => class, which isn't a very helpful message. It also catches classes found through resolver return types, not just dictionary entries, so a replacement needs an explicit check.
• SchemaParserDictionary stores a plain map since Let input types share a dictionary class #851. getDictionary() still returns a BiMap for backwards compatibility (graphql-spring-boot calls builder.dictionary(dictionary.getDictionary())), but its inverse() only keeps the first name when several input types share a class.

Proposal:

1. Internally, use two plain maps plus an explicit duplicate check that throws a SchemaClassScannerError naming the class and the types. No API change.

typeToClass.entries.groupBy({ it.value }, { it.key })
    .filterValues { it.size > 1 }
    .forEach { (clazz, types) -> throw SchemaClassScannerError("...") }
val classToType = typeToClass.entries.associate { (type, clazz) -> clazz to type }

2. Add a replacement for getDictionary() that returns Map<String, Class<*>> under a new name, and deprecate getDictionary(). Changing the return type in place would break binary compatibility.
3. Switch graphql-spring-boot to the new method.
4. Remove BiMap in the next major.