Json parser parses epoch seconds itself rather than using aws-smithy-types

[jdisanti]

The json parser currently does its own parsing for epoch seconds rather than using the code in smithy-types, and it is using a f64 to do this parsing, which can lead to loss of precision (whereas smithy-types does its own parsing into higher precision types without floating points).

The json parser should be refactored to keep a reference to the original number string as well as the parsed equivalent so that the generated token stream parsing logic can decide to parse the timestamp with the better method in smithy-types.