api_command_queue_bloc

Requires api_command_queue 0.3.x, which schedules retry backoff instead of sleeping it inside flush().

Both cubits forward the new nextDueAt, so a consumer holding a cubit or an orchestrator can tell when to flush again rather than poll for it.

Added HydratedApiCommandQueueCubit.dropUndecodableCommands, off by default.

SyncState.fromJson decodes the queue in a single pass, so one undecodable command throws and takes every other queued command with it: hydrated_bloc catches the error, falls back to an empty queue and - on its default HydrationErrorBehavior.overwrite - writes that empty queue back over the stored one. Every write the device had not yet sent is lost.

With the flag on, each command is decoded on its own first and only the ones that fail are left out, each reported through apiCommandQueueLogger. Existing behaviour is unchanged unless it is switched on.