Kwargs from DDPShardedPlugin are not forwarded to the ShardedDataParallel wrapper

## 🐛 Bug

If I pass `DDPShardedPlugin(reduce_fp16=True)` as a plugin to the Trainer, I would expect `reduce_fp16` to be sent to the `ShardedDataParallel` wrapper. This is inline with what happens in the `DDPPlugin` with the `DistributedDataParallel`.