fix: release RTP transceiver and FFI handle resources when unpublishing a track

[OlympusM]

Problem

When a local track is unpublished, some resources associated with it were never released:

• On the FFI server, the RtpSender was removed from the peer connection, but the underlying RtpTransceiver was never stopped, so it stayed alive.
• The FFI local_publication_lookup map had no corresponding cleanup, so the publication's server handle was never dropped after unpublishing, leaking the handle.

This meant repeated publish/unpublish cycles would accumulate resources that were never freed which was especially bad on a video track.

Fix

• LocalParticipant::unpublish_track now grabs the RtpTransceiver (instead of just its RtpSender) and passes it through RtcEngine::remove_track / RtcSession::remove_track, which now call transceiver.stop() after removing the sender from the peer connection.
• remove_track signatures were updated from RtpSender to RtpTransceiver across local_participant.rs, rtc_engine/mod.rs, and rtc_session.rs; the sender is now derived from the transceiver internally where needed.
• The FFI server now tracks each published track's handle in local_publication_lookup on publish, and removes + drops that handle on unpublish (room.rs), preventing the FFI-side leak as well.

Notes

• I have been testing this locally for a while now on a private application that is built on top of LiveKit.