Revise the documentation for WASI 0.3.0's tcp-socket::receive.

[sunfishcode]

receive no longer returns its own result, so update the text that says it fails if receive has already been called. Instead say that a new stream that immediately fails is returned.

And, be specific about how the stream and the future to say that either both succeed or both fail. This eliminates ambiguity about whether the stream could report a failure independently of the future.

[sunfishcode]

For additional context here, there's discussion in #111 about migrating tuple<stream<u8>, future<result<_, error-code>>> to stream<u8, result<_, error-code>> in the future, however it's likely the design of stream<T, U> would be that it would only produce a U if the stream otherwise closed successfully, which would mean that the result<_, U> could report failure when the stream itself reported success. That's unfortunate if one is able to split a stream<T> out of a stream<T, U> because it means consumers of the stream<T> wouldn't get to see that other failure.

So instead of evolving toward stream<T, result<U, E>>, we should consider evolving toward stream<T, U, E>, where U is the success value and E is the failure value. That way, if one splits a stream<T, U, E>, one would get a stream<T, _, _> which could continue to carry the full success/failure status.

This PR anticipates that evolution by specifying that we only have one success/failure condition.

[sunfishcode]

If this is accepted, I'd plan to update wasi-filesystem's corresponding read function in a similar way.

[badeend]

Looks good to me. Thanks!

---

So instead of evolving toward stream<T, result<U, E>>, we should consider evolving toward stream<T, U, E> (...)

Is this already being tracked somewhere, aside from this PR?

[sunfishcode]

So instead of evolving toward stream<T, result<U, E>>, we should consider evolving toward stream<T, U, E> (...)

Is this already being tracked somewhere, aside from this PR?

Not yet; the idea only just came up a few days ago. Luke mentioned it here and included it in his presentation here, though it's still a very rough idea at this point.