Implementation of Implicit (De-)Serializer techniques

[debasishg]

This PR offers a sample implementation of providing implicit serialization / de-serialization in Kafka Streams APIs (Issue #46).

This may not be the final implementation but provides a groundwork for further discussion. It will be great if we can have alternate implementations that make the APIs more compile time safe. An approach has been suggested by @dhoepelman in #46. But I could not make the implicit lookup work properly and it became a mess of implicits.

Design decisions adopted in the implementation :

1. The implementation offers implicits to provide the serializers and de-serializers but at the same time also allows the opt-in to use the default serializers registered in the Kafka Streams config.
2. The optional implicit pattern is implemented with the usual null-default-value trick, but with a difference. The technique used is adopted from this blog post.
3. The standard way to implement the null-default-value trick could not be applied as Scala does not allow a mix of default values and function overloads. And we have quite a few examples of such overloaded functions in the Kafka Streams API set.
4. The implementation allows implicits for the Serdes or for Serialized, Consumed and Produced. The test examples demonstrate both, though the implicits for Serdes make a cleaner implementation.
5. The implementation does a trade-off in using the null-default-value trick as it moves some of the compile time errors to runtime.

Examples:

1. The example StreamToTableJoinScalaIntegrationTestImplicitSerdes demonstrates how to use the technique of implicit Serdes
2. The example StreamToTableJoinScalaIntegrationTestImplicitSerialized demonstrates how to use the technique of implicit Serialized, Consumed and Produced.

[blublinsky]

LGTM,
very nice

[deanwampler]

Added one question about testing the usage patterns. Otherwise, LGTM

[deanwampler]

I'm not sure if I saw this, but do the contests explicitly confirm all three usage patterns. Specifically, I'm not sure I saw that #1 was being tested.

[debasishg]

You are correct! I added that test ..

[deanwampler]

Great!

[seglo]

I love the Perhaps pattern. LGTM. 👍

[seglo]

Awesome find! 👍

[notxcain]

It probably should be made final and extend AnyVal to reduce memory allocation and improve inlining.

[debasishg]

Yeah true, but in the latest release we have got rid of this abstraction altogether.