Operations which fail due to network transport errors are not retried

When a connection fails due to an IOException on send or receive, all in-flight operations on the connection are marked as failed with status code TransportFailure.

However, they are not marked with an exception which is retryable. Instead, a RequestCanceledException is thrown and they are logged as orphaned.

It seems reasonable to mark them as orphaned, as they were in fact orphaned by the first connection. But why not retry them on a fresh connection if we're within the retry timeout?