Optimization: Lazy date/timestamp parsing

[stephenh]

In theory if we did:

• Load row 1 from the database with date=2020-01-01
• Postgres sends over all types as strings
• We could delay/skip parsing the "2020-01-01" string into a Date/PlainDate until the user actually asked for that field
• We could probably even have entityResolver realize that, if the pg string encoding happens to match our GQL wire format encoding, we could go pg-string -> graphql-output also without going through a Date/PlainDate instance
• Could we do the same with fields as basic as ints? 🤔