Streams have to be counted to enforce the max stream limit. We don't
advertise a max stream limit in the TLS handshake yet, so this never
worked for IETF QUIC. Removing this will make implementing
unidirectional streams easier.
This is a lot of duplicate code for now, but it will make moving towards
the new stream ID mapping in IETF QUIC (and unidirectional streams) much
easier.
streamsMap.Range was only used to pass the stream flow control window to
all open streams after receiving the peer's transport parameters. It's
easier to pass the transport parameters to the streamsMap directly.
By introducing a callback to the stream, which the stream calls as soon
as it is completed, we can get rid of checking every single open stream
if it is completed.
The queue holds all streams that have data to send. When stream.Write or
stream.Close are called, a stream is added to this queue. It is removed
from the queue when all available (at that moment) data was sent.
This way, we don't need the round robin scheduling (which, for every
packet sent, asked every single open stream if it had data) any more.
This replaces version.UsesMaxDataFrame by versoin.UsesIETFFrameFormat.
That way, we can have two separate code paths in the unpacker to unpack
either gQUIC frames or IETF frames.
The maximum number of incoming streams doesn't need to be negotiated. It
is a value that is only announced by the peer, and has to be respected.
Furthermore, Chrome doesn't seem to care about the MSPC value anymore
(since MIDS was introduced), so there's no need to send this in the
handshake any more.
There were several bugs here:
- We must always return an error when the peer tries to open a stream
from the wrong side.
- We must never return an error when GetOrOpenStream is called for a
stream that was already closed. GetOrOpenStream is called when a frame
on a stream is received, so this happens when we receive retransmissions
of STREAM frames for a closed stream. We only used to do that check for
peer-initiated streams, not for streams we opened ourselves.
