Investigate flow control based on a dynamic window size

[bentoi]

          I wonder if control flow is the limiting factor here, see https://github.com/dotnet/runtime/issues/53372 and the linked issue https://github.com/dotnet/runtime/issues/43086

Seems similar to what we experience with ice(tcp) vs icerpc(slic)

If you capture the gRPC traffic for our comparative benchmarks, you will see the Widow size is dynamically increased.

Originally posted by @pepone in https://github.com/icerpc/icerpc-csharp/issues/3266#issuecomment-1578288963

[bentoi]

See also https://github.com/icerpc/icerpc-csharp/issues/3316