feat(session): subscribe only to the slot's own camera and mic

Every plugin source connected with auto_subscribe on, so each one pulled
every camera in the room and discarded all but one. In the 2026-10-04
old-gods-of-appalachia show (8 sources, 7 cameras) each source carried
4-7 Mbps and all 8 hit congestion in the same instant: they share the
director's single downlink, so the SFU starved the cameras on screen.

The session now connects with auto_subscribe off and calls
setSubscribed(true) on exactly the wanted publications: at connect, on
onTrackPublished (re-sweeping the participant, since SDK 1.10.1 delivers
that event's publication null), and after a reconnect. Any subscription
it did not ask for is reported and handed back. Audio-only sources no
longer subscribe-then-disable video; they never subscribe to it.

The integration test gains a bystander camera that must never be
subscribed (fails with 3 unwanted subscriptions before this change) and
a full publisher leave/rejoin as a new participant session.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

core/include/stplugin/session.h

@@ -44,11 +44,9 @@ struct SessionConfig {
 
     bool subscribe_audio = true;
 
-    /// False for an audio-only source (the soundboard, say): the wanted
-    /// video track is never attached (no AttachVideo command posted), and
-    /// its publication is explicitly disabled server-side (RemoteTrack-
-    /// Publication::setEnabled(false)) so the SFU stops sending it at all --
-    /// not just "decoded and discarded here", genuinely not delivered.
+    /// False for an audio-only source (the soundboard, say): the slot's
+    /// video track is never subscribed to (see shouldSubscribe), so the SFU
+    /// never sends it at all -- not just "decoded and discarded here".
     bool subscribe_video = true;
 
     /// How long connect() waits for the room to come up before giving up.