This would allow a "parsed" SDL file to be put into an external cache say.

All of the AST nodes are serializable - why not the parsed container of them? Hence this PR.

However in practice is not that much faster than parsing really. So caching the SDL String and reparsing might be in the same ballpark and a String is a simpler contsruct to cache

The benchmark shows that serialization is faster - YMMV

Benchmark                                                                                                    Mode  Cnt   Score   Error  Units
TypeDefinitionParserVersusSerializeBenchMark.benchMarkParsing      thrpt   15  16.590 ± 0.320  ops/s
TypeDefinitionParserVersusSerializeBenchMark.benchMarkSerializing  thrpt   15  20.066 ± 0.458  ops/s

In a low CPU constrained environment, like Serverless, this might be a faster way to pre-compute part of the schema build process.

See #2452