Track failed playback commands with bounded retry backoff

[msgerbush]

Summary

This is the first PR in the Sonos stale-IP recovery stack (refs #258). The original issue was triggered when the selected Sonos speaker changed LAN IPs, leaving the running process with a stale SoCo connection. A later pause() call then failed with No route to host.

This PR does not re-resolve stale Sonos speakers yet. Instead, it makes playback command failures safe in the domain state machine:

• Add PlaybackCommandRetry to PlaybackSession
• Only update session state after play, resume, pause, or stop succeeds
• Retry failed playback commands with bounded backoff instead of every loop tick
• Preserve retry state across brief missed tag reads
• Avoid setting paused_at when pause() did not actually pause playback

Behavior

When a playback command raises PlaybackError, the session now records retry metadata and leaves the logical playback state unchanged. Matching actions are retried after backoff; unrelated actions can clear or replace the retry state.

This prevents failed commands from making the app believe Sonos changed state when it did not.

Notes

• I think this may need to be revisited when we refactor the main event loop (Define the responsibilities in the HandleTagEvent use case more clearly #124).
• If the retry loop is exhausted trying to execute a command (e.g., pause), then the jukebox state will be stuck until another event is initiated