Refactor http, grpc senders and promote to public API

[jack-berg]

Resolves #6718.

Sketching out all the changes I think would be required to promote the HttpSender, GrpcSender interfaces to the public API. There's a lot. Summary:

• We need to hide Marshaler and related APIs, since they balloon the API surface area and would take lot of work to get ready for public. Introducing narrow focused GrpcMessageWriter, HttpRequestBodyWriter to serve the function currently performed by Marshaler.
• Need to get rid of MarshalerServiceStub and related APIs from the gRPC senders stuff. It drags in a bunch of unnecessary cruft and io.grpc.stub dependency.
• Need to hide the HttpSenderConfig#getExportAsJson() option. Senders don't need to be burdened with understanding whether the request is binary or JSON. They just need to be told the content type and a way to write the request body.
• Compressor and related APIs need to get promoted as well.
• Need to refine GrpcSenderConfig, HttpSenderConfig
  • Need to be interfaces instead of autovalue classes
  • Need javadoc
  • Make endpoint URI instead of string
  • Provide grpc senders the fully qualified service name and method name
  • Hide the Object GrpcSenderConfig#getManagedChannel(), which is required for backwards compatibility and for UpstreamGrpcSender, behind an internal-only ExtendedGrpcSenderConfig

Leaving as a draft because:

1. there are still some things to clean up
2. now that i've laid out what's necessary, I want to confirm that we want to do this
3. if we do indeed want this, I want to discuss the best way to land it. 1500 lines changes across 94 files is obviously a lot to review. I can break it into smaller parts, but we'd want to get those parts merged in one minor release to minimize churn.