split # comment descriptions on carriage returns in SchemaPrinter - #4511

Spock, SchemaPrinter with descriptionsAsHashComments(true) on a type whose description is "x\rtype Injected { a: String }", then re-parsing the printed sdl:

!reparsed.getType("Injected").isPresent()
||        |                   true
||        Optional[ObjectTypeDefinition{name='Injected', ... fieldDefinitions=[FieldDefinition{name='a', ...}]}]

printComments splits the description on \n only before writing each line as a # comment, but the lexer's Comment token also stops at \r (SourceCharacterWithoutLineFeed in GraphqlCommon.g4), so everything after a carriage return is parsed as sdl. The description can come from a \r escape in sdl or from a remote introspection result fed through IntrospectionResultToSchema and SchemaGenerator, so whoever controls a description can add definitions to the printed schema.

The hash comment branch now splits on \r\n, \r and \n. Block string and single line descriptions are unaffected: there the CR stays inside the block string or is escaped by escapeJsonString. The regression test in SchemaPrinterTest fails before the patch.