Backup codec publish robustness follow-ups (from #1159 review)

[hiroshihorie]

Deferred items from the #1159 review, none urgent since #1159 removed the paths that made the first two reachable.

• No rollback when backup codec publish fails mid flight. addSimulcastTrack inserts into simulcastCodecs before createSimulcastTransceiverSender, a failure leaves the entry with a null sender and the codec will not retry until unpublish or full reconnect. Same structure exists in JS.
• addSimulcastTrack throws on duplicate where JS logs and skips. Porting JS behavior means a nullable return (breaking API per dart-apitool) or an idempotent skip, needs a small design decision and its own test.
• Consider restartTrack() during rePublishAllTracks like JS, to recover capture after device sleep.
• Integration test that unpublish and full reconnect actually invoke clearSimulcastState (needs mock e2e container video publish support).

Cross SDK findings worth reporting upstream while porting this cleanup:

• JS gates its simulcast cleanup on trackSender non null and PC state, so the stale state fixed by Clear simulcast codec state on unpublish and full reconnect #1159 is reachable there too (LocalParticipant.ts unpublishTrack).
• Android never clears simulcastCodecs at all.
  Both fail silently because their duplicate handling skips instead of throwing, the backup codec just never republishes.