Avoid using two separate network protocols to simplify deployments behind layer 7 load balancers

[edigaryev]

More users are deploying Orchard Controller in environments with a "magical" load balancer that works at first glance, but then at some point causes issues for our gRPC reactivity & streaming connections, while the REST API keeps working, which causes confusion.

Creating a separate issue for easier discoverability and to track the progress here.

What's needed:

• figure out if gRPC is really an issue here and whether it can be re-configured somehow to avoid moving to a separate protocol
• if gRPC is indeed represents a problem in our use-case, figure out another protocol to use (e.g. ConnectRPC, WebSocket or just raw HTTP/2 streams)
• provide a backwards-compatible solution
  • we can probably just support both protocols for a while, and then at some point deprecate/remove gRPC support

See #102 for more details on the issue.