#8 ♻️ Restructure JSON deserialisation to enable other implementations than Circe

[jwojnowski]

I've extracted Circe decoders to an external module via a thin abstraction layer of EntityDecoder as a preparation for #8.

TODO

• Docs update (additional JSON module, necessary import, changelog)
• Moving tests strictly related to decoding to the Circe module
• jsoniter-scala module

Open points:

1. Scope Audience currently has an encoder, but I couldn't find any usages. I've removed it in this PR, but would be good to double-check whether that's OK.
2. There are tests of the main module (oauth2) which depend on the decoders. I can see three options here:
   1. Making the oauth2 % test depend on oauth2-circe % compile, which depends on oauth2 % compile. This might not strictly be a dependency loop, but still looks a bit odd and risky. Fortunately, SBT resolves that issue by failing with a StackOverflow (a hidden one, naturally).
   2. Separating the core module for decodable entities (done in this PR), with JSON-related modules relying only on core, allowing main tests to depend on one of the JSON modules.
   3. Creating a separate module for tests, allowing them to depend on oauth2 and one of the JSON modules without a loop.

[majk-p]

Regarding the open point on the Audience encoder - let's drop it. This library is intended as a client for OAuth2 providers, thus it only needs to understand how to decode the data, it should not need the encoder. Let's drop it for now and see if it happens to be a problem when early adopters attempt to use RC version.

[majk-p]

LGTM and publish as an RC