feat(session): round out accept-side request helpers

[floatdrop]

What

The accept side of the request API was asymmetric: AcceptSubscribe returned a rich *Publication, AcceptPublish returned a bare error, and Fetch / TrackStatus / the three namespace requests had no Accept* helper at all (servers hand-rolled req.Reply). This rounds the set out so every outbound opener has a matching, typed accept-side counterpart.

Outbound	Accept-side (new/changed)	Handle
Subscribe → *Subscription	AcceptSubscribe (unchanged)	*Publication
Publish → *Publication	AcceptPublish	*IncomingPublication (TrackAlias, Update)
Fetch → *FetchRequest	AcceptFetch	*FetchResponder (OpenFetchStream, Request ID pre-bound)
TrackStatus → *TrackStatusRequest	AcceptTrackStatus	— (one-shot, documented)
PublishNamespace → *NamespacePublication	AcceptPublishNamespace	*IncomingNamespacePublication
SubscribeNamespace → *NamespaceSubscription	AcceptSubscribeNamespace	*IncomingNamespaceSubscription
SubscribeTracks → *TrackSubscription	AcceptSubscribeTracks	*IncomingTrackSubscription (WritePublishBlocked)

Highlights:

• AcceptPublish now returns *IncomingPublication (was bare error) — the receiving side of a publisher push, with TrackAlias() and Update().
• AcceptFetch writes FETCH_OK and returns a responder whose OpenFetchStream() binds the fetch's Request ID onto the FETCH_HEADER uni-stream automatically — mirroring how AcceptSubscribe's Publication.OpenSubgroup pre-binds the alias.
• AcceptSubscribeTracks' handle has WritePublishBlocked, the publisher-side mirror of TrackSubscription.ReadPublishBlocked.
• AcceptTrackStatus deliberately returns no handle (TRACK_STATUS is a one-shot status query with no object-push side) — documented, not left as an arbitrary gap.

The Incoming* names follow the package's existing convention for the receive/accept side (IncomingSubgroupStream, IncomingFetchStream).

Testing

• New accept_test.go: round-trips for each helper, including a full FETCH OpenFetchStream exchange (asserting the bound Request ID) and a WritePublishBlocked → ReadPublishBlocked exchange. The streaming tests run the server side in a goroutine and read on the client in send order, since the sessiontest pipes are synchronous (a deadlock here was caught by go test ./... and fixed).
• Updated the AcceptPublish callers/example for the new signature; added a README row.
• go build ./..., full go test ./... (every package ok), golangci-lint run, and the modernize check — all clean.

🤖 Generated with Claude Code