[runtime/network] Allow read timeout recovery

[clabby]

Overview

Currently, if a read times out in the tokio Stream impl, the partially read data will be lost due to read_exact not being cancel safe. We currently disconnect upstream of this code if a timeout occurs, but if we want to support recovery, we'll need to consider holding read data in a buffer (or in an error variant) for recovery purposes.

monorepo/runtime/src/network/tokio.rs

Lines 37 to 52 in 2fbf479

	impl crate::Stream for Stream {
	async fn recv(&mut self, buf: impl Into<StableBuf> + Send) -> Result<StableBuf, Error> {
	let mut buf = buf.into();
	if buf.is_empty() {
	return Ok(buf);
	}
	// Time out if we take too long to read
	timeout(self.read_timeout, self.stream.read_exact(buf.as_mut()))
	.await
	.map_err(|_| Error::Timeout)?
	.map_err(|_| Error::RecvFailed)?;
	Ok(buf)
	}
	}