Add a new type of Serializer for supporting zero allocation scenarios

[nmosafi]

Right now ISerializer and ISerializerAsync are forced to allocate a new array for every message which is sent. For a high throughput application we'd like to avoid allocating large amounts of memory.

The API exposed by librdkafka supports buffer reuse as it accepts a length and offset alongside each byte array, however this is not exposed by the serialization framework in Kafka dotnet.

The simplest approach here is to allow ArraySegment<byte> to be returned by the serializer rather than byte[], however changing all serializers to return ArraySegment<byte> would surely be a large breaking change. So instead I decided to implement this as a new type of serializer.

Please give me some feedback and thoughts on this PR.

Thanks

[cla-assistant]

All committers have signed the CLA.

[nanov]

Is there a reason why this is not looked at, such functionality could allow, potentially, huge performance benefits in scenarios where the implementations relays on pooled arrays ( only minimum length is guaranteed, array may be bigger depending on the pool ) or ring buffers.