Transport: Partial reliability

[aboba]

Partial reliability is an attractive addition to frame/stream transport. By setting a timer before calling writer.write(), it is possible to give up on a "discardable frame" (a non-base layer frame) after it has taken too long to send.

Live Demo: https://webrtc.internaut.com/wc/wtSender10/
Code: async function writeChunk(transport, chunk, rto, info)

The issue is that RESET_STREAM may not be uniformly implemented in QUIC stacks. RFC 9000 Section 19.4 says:

"After sending a RESET_STREAM, an endpoint ceases transmission and retransmission of STREAM frames on the identified stream. A receiver of RESET_STREAM can discard any data that it already received on that stream."

Note the lack of normative language with respect to receive-side processing, where RFC 9000 Section 3.2 does not require that RESET_STREAM be immediately surfaced to the application.

The handling of RESET_STREAM frames also affects relays, which may immediately process a RESET_STREAM frame on receipt, forwarding it on and not forwarding subsequently received data, or alternatively, could forward incoming data in the order it was received, including RESET_STREAM frames.

These issues appear to be affecting the demo application. RESET_STREAM frames are sent when the RTO timer expires, but even when writer.abort() is called, frames typically arrive intact anyway.

Related: w3c/webtransport#422