[Schema Registry] Explore caching in SchemaRegistryClient

[ramya-rao-a]

This issue is a follow up for #10437

Once we have a convenience layer over the auto-generated SchemaRegistryClient, the next step would be to support caching for some of the operations on the client.

This issue is to

• identify which operations need caching
• get guidelines around whether caching should be done at the client level or the consumer of the client is responsible for this
• if this is done at the client level, what should be in the "raw response" section of the response of the cached operations

[nguerrera]

Now that I've implemented some caching in the serializer, I am more convinced that caching at the client is the wrong layer. When we cache in the serializer, we can cache by the schema content directly. However, before we can call the client, we need to parse the schema to get its name. This would mean an expensive parsing step on every serialize that can be saved by moving the cache up one layer. And it also eliminates all of the questions about caching responses or not. I think we should keep the cache in the serializer, but explore how to configure its policy. (For the initial preview, you will have to throw the serializer away and create a new one to free the entries in the cache as discussed elsewhere.)