WebRTC media transport for voice/video

## Summary
Implement WebRTC media transport for voice and, later, video features once signaling semantics are settled.

## Scope
- Define the browser transport path for realtime media in supported web-desktop environments.
- Support the core transport lifecycle: negotiation, reconnect, teardown, and device changes.
- Align the transport with the product rules defined by voice/session/permission issues.
- Keep the initial scope focused on reliable voice transport before broader video parity.

## Acceptance Criteria
- [ ] Voice media can establish, recover, and tear down cleanly in supported browsers.
- [ ] Device changes and reconnects do not leave ghost sessions behind.
- [ ] Transport failures produce visible recovery or fallback states.
- [ ] The implementation leaves room for later video and screen-share expansion.

## Operational Notes
- Self-hosted deployments need explicit assumptions about NAT traversal, TURN requirements, and bandwidth cost.
- This issue should not pretend video parity is free once voice works.

Start by resolving the signaling semantics and reviewing the related voice, session, and permission issues. Then define the supported browser and self-hosted assumptions, including NAT traversal and TURN requirements. Done means reliable voice negotiation, recovery, teardown, device changes, and visible fallback states without ghost sessions.

Written by the indexing model from the issue text.