soroban-rpc: fix datarace in bufferedResponseWriter.WriteOut

[tsachiherman]

What

Fix data race #959 .

Why

When implementing bufferedResponseWriter.WriteOut, I attempted to implement the writing to the output stream asynchronously, so that in case of context expiration the writing to the channel would be suspended at once.

Unfortuntally, it led to conflict with go http server library : When the bufferedResponseWriter.WriteOut could not complete writing to the stream in time, the method would return. However, the writing operation was still going on it's own pace.
then, the http server was attempting to write to the stream, creating a data race.

The solution is to write to the stream in smaller chunks, and continuously check the ctx.Err in order to examine when it expires.

Known limitations

n/a

[2opremio]

I would advocate to simply writing all at once. I think responses are small, so I think we can afford to surpass the deadline by the writing time.

[tsachiherman]

I would advocate to simply writing all at once. I think responses are small, so I think we can afford to surpass the deadline by the writing time.

I'd like to keep it for protection against malicious client leaving the connection open and draining it at a low rate.

[2opremio]

Chunks are not going to help you with that either, they can block when reading a single byte

And the MTU doesn't help in all cases, making an assumption on the link layer is overkill at this level of abstraction