fix: stop camera streams from ducking system audio

[mvanhorn]

Summary

Camera streams no longer duck (lower the volume of) other system audio. The stream players now configure their audio session so opening a camera view stops interrupting music, calls, or other playback, and the session is deactivated cleanly when the stream closes.

Why this matters

Issue #138 reports that viewing a camera ducks system audio: starting an HLS or WebRTC camera stream activated an audio session with default options, so iOS lowered other audio even though these streams are typically muted video. Users watching a camera had their music or podcast volume dropped for no reason.

The stream clients now use a non-ducking audio-session configuration and balance activation with deactivation on teardown. Because LKRTCAudioSession (WebRTC) and the AVAudioSession are reference-counted, activation and deactivation are guarded by a per-client didActivateAudioSession flag so each client activates at most once and deactivates exactly once. deinit routes through the idempotent teardown, so normal disconnect(); client = nil flows and stream reopen do not drive the reference count negative or leave the session active.

Testing

Verified the activation/deactivation is balanced and idempotent by tracing the disconnect and deinit paths and the mute path (which clears the flag after deactivating). On-device build/run was not exercised in this environment (no iOS simulator available); the change is scoped to audio-session setup and teardown in the two stream clients.

Fixes #138

[nickustinov]

Thanks Matt! Reviewed and verified – activation/deactivation is balanced on every path (connect/disconnect, play/stop, mute toggle, fallback chain, deinit), and the build passes on Mac Catalyst. Nice clean fix for #138.

One note for the future: since AVAudioSession.setActive(false) isn't reference-counted, the unconditional deactivation in both clients would need revisiting if we ever add a multi-camera live grid. Fine as-is with the single-stream architecture.