Add use_generic_streaming_requests option to make testing server impls easier

[yotamofek]

Motivation

Currently, testing the server trait implementations of client-streaming methods (i.e. requests that receive a stream) is quite difficult, and requires jumping through hoops in order to construct a Streaming object (see #462 for one way of doing it).

Solution

Adds a codegen option that makes the server trait go from this:

trait Test {
    async fn test_request(
        &self,
        req: Request<Streaming<Message>>,
    ) -> Result<Response<Message>, Status>;
}

to this:

trait Test {
    async fn test_request(
        &self,
        req: impl IntoStreamingRequest<Message = Result<Message, Status>, Stream: Unpin> + Send,
    ) -> Result<Response<Message>, Status>;
}

Since Request<Streaming<Message>> impls IntoStreamingRequest<Message = Result<Message, Status>>, the grpc server can continue invoking these handlers in the same way, but it also allows any object implementing Stream<Item = Result<Message, Status>> to be used.
This makes testing much easier.

[yotamofek]

Just realized this uses associated type bounds which were only stabilized in 1.79.0.
How often is the MSRV raised?
The Unpin bound in IntoStreamingRequest<Stream: Unpin> isn't strictly necessary (and might actually make the trait less flexible), and can be either removed or emulated with a named generic param (i.e. test_request<S: Unpin>(req: impl IntoStreamIterator<Stream = S>) or so)